Seminars

Mobility plays a vital role in our daily lives, encompassing activities such as commuting to and from work, running errands, and taking vacations. Individuals often display similar mobility behaviors when sharing everyday responsibilities. For example, commuters typically travel to work during rush hour, parents need to drop their kids at school, and university students tend to go out on Friday nights. Recent quantitative studies have identified a gender gap in human mobility, revealing distinct behaviors between men and women. Women tend to walk shorter distances, visit fewer locations and spend more time at home. However, the factors driving these differences remain unclear. In this talk, we will demonstrate how smartphone data can be utilized to unveil gender gaps in human mobility. First, we will present the findings from a unique data set containing GPS traces of over 250,000 individuals, exposing the consistent gender gap in mobility across countries, age groups, and urbanization levels. Second, we will explore the influence of employment and investigate whether the gender gap in the labor market might explain the observed gender gap in mobility. Third, by leveraging smartphone application data, we will examine how becoming a parent transforms the mobility patterns of males and females. Lastly, we will use a city-level physical activity dataset to investigate how the urban environment shapes mobility and contributes to gender differences. Through this session, I hope to foster an open discussion on how we can continue researching the role of gender, race, or socio-economic inequalities in human mobility. Bio: Silvia de Sojo Caso is a Ph.D. candidate in the Section for Cognitive Systems at the Technical University of Denmark, supervised by Laura Alessandretti and Sune Lehmann. She received a BSc in Industrial Engineering in 2016 from the Polytechnic University of Catalonia (UPC), Spain, and an MSc in Business Analytics in 2022 from the Technical University of Denmark (DTU). Her professional experience includes various roles, from Data Analyst to Project and Program Manager at HP Inc, Barcelona, and UNDP, UN City Copenhagen. Her research focuses on understanding behavioral inequalities across genders and diverse groups. She analyzes large-scale data sets, including GPS traces from over 6 million users, smartphone app usage, and online user-generated data. She uses a cross-disciplinary approach to integrate methodologies such as network analysis, distributed computing, and matching techniques. In her Ph.D., she explores the gender gaps in human mobility across countries, cities, and socio-economic groups. Using causal inference methods, she examines the influence of household responsibilities, division of labor, and safety perceptions on gender differences in mobility.

The growing ubiquity of deep neural networks in daily life highlights the need to address the challenges and limitations they face when deployed on edge devices. These devices impose constraints on resources, accessibility, and scalability, and are subject to dynamic changes in their working environment. This makes the optimization and generalization of neural networks a critical area of research. In this talk, we adopt an empirical science approach, treating deep learning as an observable and experimental phenomenon. Our primary goal is to understand when neural networks work and when they do not, and we focus on examining the optimization and generalization of neural networks in the context of edge environments.

While digital technologies are an important factor in modern business transformations, digital entrepreneurship shows to be a multi-faceted phenomenon that interconnects humans, technology, and the environment. This determines multiple links between different systems, positive and negative feedback, emergent system patterns and adaptive behavior, and connections across space and time. In this research, we aimed to investigate how the changing business patterns, entailed by digital entrepreneurship, are embedded into a broader context of innovation systems influencing social cohesion, economic growth, and innovation development. The cultural aspect was incorporated to build a link to individual perspectives of entrepreneurs, referring to innovation attitudes, decision-making process and effectuation, competencies, and adaptive capacities. Therefore, the differentiating features of migrant and native entrepreneurs were additionally considered. For analyzing this complex real-life phenomenon, an iterative system modeling based on causal loop diagrams was conducted. Further, a transdisciplinary approach was applied as part of the empirical research in order to accumulate knowledge of science and practice during the modeling process.

This research examines how production chains and networks of clients impact the emergence of pioneering industries in a particular location. The study utilizes a unique and comprehensive private dataset containing detailed information on registered firms and individuals in Ecuador, including transaction records and other characteristics. Various relatedness indicators, existing and newly proposed ones that consider supplier and client networks, are calculated using this dataset. Logistic regressions and machine learning techniques are employed to assess the likelihood of a pioneering industry emerging as an indicator of diversification. The results reveal that the principle of relatedness based on co-location holds in an oil exporting economy, and the network of suppliers is more influential than the network of clients in the emergence of new economic activities. However, the relevance of these variables varies depending on the industry, and the presence of oil and mining-related activities hinders local economies’ diversification efforts.

The study of collective synchronous behavior has primarily focused on the analysis of abstract and greatly simplified mathematical models. Many applications of these models to living systems have been proposed, but the incorporation of real-world data is unfortunately rare. I will present new data and analysis regarding synchronization phenomena observed in one species of firefly from southeast Asia. Due to its relative immobility during synchronous flashing displays, this species offers a unique opportunity for reliable tracking and direct application of candidate models. In late 2022, we used stereo videography to document the three-dimensional behavior of multiple swarms over multiple nights. Our results show that swarms exhibit “meta oscillations” characterized by order parameters that rise and fall on an intermediate time scale (~40 times longer than the typical flashing period of a firefly). This is consistent with models suggesting a “breathing” chimera state—a unique type of spatiotemporal organization that has been the subject of extensive theoretical study, but which has rarely been observed in nature. In addition, I will present new directions that we are now taking with further recordings, analyses, and potential experiments.

Cooperation is a reciprocal behavior that benefits others while costing oneself. In this work, we propose a general mathematical framework to investigate the dynamics of a typical public good game that involves cooperators and defectors in a patch-based ecological system. We first use the generalized modeling method to engineer a public goods game. Then by introducing the diffusion among spatial patches, we examine that the coupling factors that evolve in the networks of patches can trigger various instabilities such as shortwave instability and finite wave number instability in three different scenarios, representing other cooperative behaviors among players. Collectively, the perspectives in this work shed new light on the intricate relationship between cooperators and defectors in public goods game, and highlight opportunities for analyzing nonlinear dynamics of public goods models in ecological systems and beyond efficiently.

The past 15 years saw the introduction of several technologies with the potential for “disruptive change” in urban transportation systems. However, the full implications of such changes are often not fully understood, with contrasting claims made by parties with opposing interests, such as technology companies, cities, taxi drivers, etc. It is thus essential to critically explore the implications of technological change. In this talk, I will first review some of the main issues about on-demand urban mobility (i.e. “ride-sourcing”) and then present our recent research focusing on one crucial aspect: the trade-off between achieving competition and coordination among mobility operators. Our research shows that compared to a scenario with central coordination, having a segmented mobility market with competing operators can have significant drawbacks by increasing combined fleet size requirements. I will discuss some policy implications for cities of different sizes and travel demand.

Empirical testing of the income-enhancing effect of technology diffusion has been traditionally carried out using aggregate measures of country openness as a proxy of the extent to which a country is exposed to foreign markets and migration flows. For example, to account for countries’ exposure to international trade, it is customary to employ a simple measure of trade openness computed as the ratio between a country’s total trade and its gross-domestic product (GDP). However, these measures are essentially local, as they only account for direct interactions with neighboring countries. Indeed, it may be the case that such openness proxies are not perfectly correlated with indicators accounting for the global embeddedness of a country in the networks of international relations, which instead fully account for the overall position of a country within the complex web of interconnections in which it is entrenched. If that is the case, standard measures of openness are not able to capture the income-enhancing effects of international technology diffusion, which may be instead better proxied if one employs tools and concepts borrowed from complex-network theory. In this work, we present indeed a simple theoretical country-growth model predicting that, net of country-specific spatio-temporal characteristics (including traditional openness proxies), country per-capita income should positively depend on its global importance (i.e., by her Bonacich or Katz centrality) in the macroeconomic networks wherein she is embedded. Next, we take to the data the implications of the theoretical model, using data on the international networks of merchandise trade, finance, migration and ideas’ flows. We build a multi-layer network and we employ different measures of country centrality in such a network as a covariate in panel regressions explaining country per-capita income. The empirical exercises strongly support the predictions of the model, robustly across a number of alternative specifications of the empirical model and controlling for possible endogeneity issues and spatial effects.

Psychology is critical to understanding human history. When aggregated, changes in people’s individual preferences can lead to important changes in institutions, social norms, and cultures. Studying the role of psychology in shaping human history has been hindered by the difficulty in recovering the thoughts and preferences of people who are no longer alive. Recent developments in psychology suggest that cultural artifacts reflect, in part, the psychology of the individuals who produced or consumed them. Cultural artifacts can thus serve as “cognitive fossils,” i.e., physical imprints of the preferences of long-dead people. In this talk, I will discuss a series of studies suggesting the relationship between political revolutions in the early modern period and secular changes in peoples’ preferences as expressed in cultural artifacts. I will also discuss how these cultural artifacts can reveal changes in cognition and dominant moral virtues in periods of scientific progress. Hinging on these insights, I will discuss how applying similar tools to modern culture can reveal long trends in psychological preferences and how these can help us understand current social and political dynamics.

Poverty maps are essential tools for governments and NGOs to track socioeconomic changes and adequately allocate infrastructure and services in places in need. Sensor and online crowd-sourced data combined with machine learning methods have provided a recent breakthrough in poverty map inference. However, these methods do not capture local wealth fluctuations, and are not optimized to produce accountable results that guarantee accurate predictions to all sub-populations. Here, we propose a pipeline of machine learning models to infer the mean and standard deviation of wealth across multiple geographically clustered populated places, and illustrate their performance in Sierra Leone and Uganda. These models leverage seven independent and freely available feature sources based on satellite images, and metadata collected via online crowd-sourcing and social media. Our models show that combined metadata features are the best predictors of wealth in rural areas, outperforming image-based models, which are the best for predicting the highest wealth quintiles. Our results recover the local mean and variation of wealth, and correctly capture the positive yet non-monotonous correlation between them. We further demonstrate the capabilities and limitations of model transfer across countries and the effects of data recency and other biases. Our methodology provides open tools to build towards more transparent and interpretable models to help governments and NGOs to make informed decisions based on data availability, urbanization level, and poverty thresholds.

As urbanization continues to grow, cities face increasing pressure to provide efficient, affordable, and environmentally-friendly transportation options. We will explore the different approaches that cities can take to promote sustainable mobility, such as the implementation of public transit, cycling and walking infrastructure.

The mathematical laws of life manifest scaling regularities such as the relationship between mass and metabolism for the smallest to the largest organisms on Earth. These laws lack essential components representing interaction between organisms while sharing limited resources. Once accounted for, these components can bring significant variation to the predicted demographic laws using just metabolic scaling theory and can give mathematical descriptions for observed ecological phenomena. The oscillations in population number, where spikes in the number of organisms of a specific size propagate from small to large organisms is an example of such a phenomena. Here, we incorporate spatial competition and resource variation in a differential equation model for the population dynamics of sessile organisms. We use analytic and numerical tools to solve the corresponding equations and to characterize the form of instabilities that generate the oscillations, which we use to identify hidden mechanisms that may drive instabilities in ecological systems such as forests. As a result, we may be able to identify the most significant factors that affect the stability of an ecosystem corresponding to resource fluctuations that may become more prominent with climate change.

A recent study by Castro Torres and Alburez-Gutierrez (2022) shows that social sciences studies using data from the global North (Europe and North America) are less likely to contextualize their title than studies from the global South. The authors conclude that this omission of geo-contextualization from the global North could potentially constitute an unwarranted claim on universality and may, in turn, lead to lesser recognition and impact of global South studies. However, they did not analyze citations to verify this hypothesis. The study “On the impact of geo-contextualized and local research in the global North and South” (Mongeon, et al., 2022) investigates the relationship between geographical contextualization in research articles and their scholarly impact. The study analyzed 29,850,298 articles and reviews published between 1997 and 2020. The results showed that less than 17% of the publications mentioned a geographical location in the title or abstract, and less than 7% were considered local research. The study found that publications with affiliations from the global South have a citation disadvantage in all fields, except in Engineering and Technology. Local research on the global South showed a reverse trend, but the global North still dominates in terms of production and impact. Following a short introduction of the study motivation and results, Named Enity Recognition (NER) as method for information extraction in Scientometric Research will be evaluated and discussed.

The focus of my talk is on distribution networks. In these networks, manufacturers and distributors work closely together to ensure consumers have access to goods and services. First, I will present a novel dataset, the ARCOS, which is still unexplored by the scientific community. This collects nearly half a billion records of opioid shipments in the United States. Then, I will describe how I leveraged these data to reconstruct and empirically analyze the opioid distribution network in the United States. Next, I will present two network-based dynamic models to investigate this system. The first model explains the system formation; the second describes its possible response to supply-side shocks. I will conclude the talk with a final overview of the main findings.

The dynamics of firms can be understood by studying the properties of their growth rates. Although there has been a significant amount of work in trying to clarify the statistical properties of the growth-rate distribution, mimicking in many ways the tradition of studying the statistical properties of financial returns, a clear picture of their inter-temporal dynamics is still missing. In this talk, I will present preliminary results on this topic. I will show that firms have an auto-correlated growth structure, but also that the structure of the correlations between firms shows a lot of interesting features –such as oscillations with non-trivial frequencies — whose origin is yet to be clearly elucidated.

Normative theories and statistical inference provide complementary approaches for the study of biological systems. A normative theory postulates that organisms have adapted to efficiently solve essential tasks and proceeds to mathematically work out testable consequences of such optimality; parameters that maximize the hypothesized organismal function can be derived ab initio, without reference to experimental data. In contrast, statistical inference focuses on the efficient utilization of data to learn model parameters, without reference to any a priori notion of biological function. Traditionally, these two approaches were developed independently and applied separately. Here, we unify them in a coherent Bayesian framework that embeds a normative theory into a family of maximum-entropy ‘‘optimization priors.’’ This family defines a smooth interpolation between a data-rich inference regime and a data-limited prediction regime. Using simple examples from neuroscience and gene regulation, we demonstrate that our framework allows one to address fundamental challenges relating to inference in high-dimensional, biological problems.

We present a minimalist agent-based model to efficiently simulate the Ethereum Proof-of-Stake (PoS) consensus protocol. The model depends on information diffusion: the main parameters are block and attestation latency as well as the underlying peer-to-peer topology. In addition to the model description, we define measures to asses the quality of consensus, providing tools to observe the model’s reactivity to parameters change. Nicolò Vallarano's fields of research range between the statistical mechanics of networks and distributed ledger technologies. He’s a research associate at UZH Blockchain and Decentralized Ledger Techonology (BDLT) group modelling consensus protocols other than the classic Bitcoin proof-of-work. As part of his research activity he collaborates to the Blockchain Observatory in the study and classification of cryptocurrencies economic state.

Cities host diverse people and their mixing is the engine of prosperity. In turn, segregation and inequalities are common features of most cities and locations that enable the meeting of people with different socio-economic status are key for urban inclusion. In this study, we adopt the concept of economic complexity to quantify the ability of locations – on the level of neigh- borhoods and amenities – to attract diverse visitors from various socio-economic backgrounds across the city. Utilizing the spatial distribution of point of interests inside the city of Budapest, Hungary, we construct the measures of amenity complexity based on the local portfolio of di- verse and non-ubiquitous amenities. We investigate mixing patterns at visited third places by tracing the daily mobility of individuals and characterizing their socio-economic status by the real-estate price of their home locations. Results suggest that measures of ubiquity and diversity of amenities do not, but amenity complexity correlates with the diversity of visitors to neigh- borhoods and to actual amenities alike. We demonstrate that, in this monocentric city, amenity complexity is correlated with the relative geographic centrality of locations, which in itself is a strong predictor of socio-economic mixing. Our work combines urban mobility data with economic complexity thinking to show that the diversity of non-ubiquitous amenities, central locations, and the potentials for socio-economic mixing are interrelated.

Traders buying or selling an asset receive price if another party observing their order jumps in front of them, i.e., front-running. These kinds of attacks are prevalent in decentralized finance. We will present a characterization of such attacks and discuss possible solutions and their feasibility.

Phosphorus is one of the key elements in the production of fertilizers and thus the production of food. Since phosphorus is constantly removed from the soil in the process of agricultural production, its reliable availability in the form of fertilizers is essential for food security and economic development. In this paper we present a new method to trace the flows of phosphorus from the countries where it is mined to the counties where it is used in agricultural production. We achieve this by combining data on phosphorus production with data on fertilizer use and data on international trade of phosphorus-related products. We show that by making certain adjustments to data on net exports we can re-construct the matrix of phosphorus flows to a large degree, a results that is important for devising measures on sustainable development and environmental accounting, not only for phosphorus but for many other resources.

Patients become increasingly multimorbid with age leading to decreased quality of life, increased need for hospitalizations, health care utilization, mortality, and care costs. However, the typical population-scale disease trajectories along which patients become multimorbid are not yet fully understood. We use a unique dataset containing 45 million records of almost all inpatient stays from 2003 to 2014 in Austria to investigate these disease trajectories. Our multilayer comorbidity network approach shows how to identify critical events that may contribute to determining a patient’s future life-course disease trajectory.

Credit risk assessment is core to the banking business. This risk is materialised when a counterparty is unwilling or unable to fulfill its contractual obligations.Traditionally, credit risk models focus on the borrowers financial conditions, by using financial statements. However, recent crises have drastically revealed that the upstream and downstream propagation of shocks along supply chains can potentially lead to firms’ insolvency. Traditional credit risk models using solely node-level financial information can not take the risk of supply contagion into account, leading to potential underestimation of credit risk losses. Here we address this issue by answering two questions. Can a single firm affect financial stability by causing large production losses in the supply chain network? How are banks exposed to the additional risks from shock propagation in the supply chain network? Our study is based on a unique country wide dataset containing all major supply chain links of Hungarian firms in combination with firm-bank loans. First, we introduce a financial systemic risk index (FSRI) of companies that measures financial losses of the banking system through direct loans and indirect financial losses caused by the default of other firms due to supply chain disruptions. We show that a small fraction of firms pose sizeable risks to the financial system, affecting up to 17% of overall bank equity. This magnitude of risk is explained mostly by propagation of shocks in the supply network and not by their direct impacts on banks. Further, we calculate expected losses, value at risk and expected shortfalls of banks with and without supply network contagion. Our simulations show that EL, VaR, and ES of banks are on average 3 times higher. Our findings show that it is crucial for regulators’ financial systemic risk assessment to monitor supply network shock propagation in order to have a more complete picture of threats to financial stability.

Bitcoin, the first implementation of the blockchain technology, is frequently abused by cyber-criminals: scams, extortion, thefts, ransomware, etc. In this talk we will present a novel, fully automated transaction tracing technique, useful for searching for financial relations between malicious actors and (benign) services, which can lead to their attribution. Together with his colleagues, Gómez evaluated back-and-forth exploration on 30 malware families. They build oracles for 4 families using Bitcoin for C&C and use them to demonstrate that back-and-forth exploration identifies 13 C&C signaling addresses missed by prior work, 8 of which are fundamentally missed by forward-only explorations. Their approach uncovers a wealth of services used by the malware including 44 exchanges, 11 gambling sites, 5 payment service providers, 4 underground markets, 4 mining pools, and 2 mixers. In 4 families, the relations include new attribution points missed by forward-only explorations. It also identifies relationships between the malware families and other cybercrime campaigns, highlighting how some malware operators participate ina variety of cybercriminal activities.

The concepts of build-up and spontaneous relaxation of “tensions” due to reorganizations within complex systems have been the defining characteristics of self-organised criticality. They have also found their places in econophysics literature. I will take this further for socio-technical systems such as transport and supply chains: using extensive real-life operational data from the Dutch railways, I will demonstrate that a temporal network is a very well-suited language for describing such tensions in the system to trace out its systemic vulnerability to spreading [1]. Touching upon “a system more optimized for efficiency is more vulnerable to spreading phenomena”, I will develop “entanglement entropy” as a measure for tension in temporal networks [2]. The framework will also bring us to a completely different application: spreading of infectious diseases (the specific application is reanalysing/modelling the first COVID-19 wave in the Netherlands for policy purposes). [1] M. M. Dekker and D. Panja. Cascading dominates large-scale disruptions in transport over complex networks. PLOS ONE 16, e0246077 (2021). [2] M. M. Dekker, R. D. Schram, J. Ou and D. Panja. Hidden dependence of spreading vulnerability on topological complexity. Phys. Rev. E 105,054301 (2022). [3] M. M. Dekker, L. E. Coffeng, F. P. Pijpers, D. Panja and S. J. de Vlas. Reducing societal impacts of SARS-CoV-2 interventions through subnational implementation. https://www.medrxiv.org/content/10.1101/2022.03.31.22273222.abstract (in revision for eLife).

This Talk intends to provide a direct understanding of the twin transition from the innovative activity domain. It will start with a technological mapping of the technological innovations characterised by both climate change mitigation and labour-saving attributes. To accomplish the task, I draw on the universe of patent grants in the USPTO since 1976 to 2021 reporting the Y02-Y04S tagging scheme, defined by the USPTO as patents referring to climate change mitigation technologies. After the identification of green technologies, by means of a textual-content algorithm, we identify those patents reporting an explicit labour-saving heuristic. I characterise their technological, sectoral and time evolutions. According to our analysis, many of the technologies addressing climate change transition, as renovation of buildings or energy efficiency, and eventual occupations involved in their use, such as installers of photovoltaic panels, are characterised by increasing labour-efficiency in their production and adoption. In particular green technologies haracterized by labour saving (LS) heuristics exert much more pronounced growth rates with respect to the no LS ones. Our findings challenge the common understanding of the ”green transition” as only labour augmenting. Potentially, the capacity of the ”green” segment as a net labour-absorber is weaker than commonly expected. Direct policy interventions are necessary beyond adaptation policies to ”green skills” currently envisaged by institutions.

Did migrants help make Paris a Mecca for the arts and Vienna a beacon of classical music? Or was their rise a pure consequence of local actors? Here, we use data on the biographies of more than 22,000 famous historical individuals born between the years 1000 and 2000 to estimate the contribution of famous immigrants, emigrants, and locals to the knowledge specializations of European regions. We find that the probability that a region develops a specialization in a new activity (physics, philosophy, painting, music, etc.) grows with the presence of immigrants with knowledge on that activity and of immigrants specialized in related activities. Similarly, we find that the probability that a region loses one of its existing areas of specialization decreases with the presence of immigrants specialized in that activity and in related activities. In contrast, we do not find robust evidence that locals with related knowledge play a statistically significant role in entries or exits. Together, these findings advance our understanding of the role of immigrants, emigrants, and locals in the historical formation of knowledge agglomerations.

In a rapidly changing world, facing an increasing number of socioeconomic, health and environmental crises, complexity science and data science can help us quantify vulnerabilities and monitor progress towards achieving the UN Sustainable Development Goals. In this talk, she will provide a non-exhaustive overview of the main areas of applications where data-driven computational methods have shown their potential for social impact. She will then give a more deep-dive, specifically into her work on predicting food insecurity from conflict, weather, and economic data.

Polyomino shape is a union of unit squares connected edge by edge. In mathematical physics, they are also known as animals in the grid. We define a simplicial complex for a given set of polyomino shapes and a given subset of the square grid in a plane or on a torus. The topological and combinatorial properties of the complex are exciting. They reveal many features of placements of tiles from the given set into the given subset of the grid without overlappings. We will prove that these complexes have high connectivity and that, in some cases, they have the homotopy type of wedge of spheres. Some of their unexpected applications will be discussed.

There exist a relative lack of sophisticated methods to compare the network topology of networks. Here we propose a new method to compare the modular structure of a pair of node-aligned networks. The majority of current methods, such as normalized mutual information, compare two node partitions derived from a community detection algorithm yet ignore the respective underlying network topologies. Addressing this gap, our method deploys a community detection quality function to assess the fit of each node partition with respect to the other network’s connectivity structure. Specifically, for two networks A and B, we project the node partition of B onto the connectivity structure of A. By evaluating the fit of B’s partition relative to A’s own partition on network A (using a standard quality function), we quantify how well network A describes the modular structure of B. Repeating this in the other direction, we obtain a two-dimensional distance measure, the bi-directional (BiDir) distance. The advantages of our methodology are three-fold. First, it is adaptable to a wide class of community detection algorithms that seek to optimize an objective function. Second, it takes into account the network structure, specifically the strength of the connections within and between communities, and can thus capture differences between networks with similar partitions but where one of them might have a more defined or robust community structure. Third, it can also identify cases in which dissimilar optimal partitions hide the fact that the underlying community structure of both networks is relatively similar. We illustrate our method for a variety of community detection algorithms, including multi-resolution approaches, and a range of both simulated and real world networks. Biography: Neave O’Clery is Associate Professor and Director of Research at the Centre for Advanced Spatial Analysis (CASA) at University College London where she leads an inter-disciplinary research group focused on network and data-driven models for economic development and urban systems. She is also a Turing Fellow at the Alan Turing Institute, as well as a Visiting Fellow at the Oxford Mathematical Institute and an Oxford Martin Fellow. Her work spans a number of topics and fields including structural change and industrial development, economic complexity and evolutionary economic geography, the informal economy, urban mobility and segregation, and network science. She also works alongside a number of policy and government institutions ranging from city majors to global multi-laterals including the Greater Manchester Combined Authority, the Irish Department for Enterprise, Trade and Employment, and the World Bank. Neave was previously a Senior Research Fellow at the Mathematical Institute at the University of Oxford, and before this a Fulbright Scholar and Postdoctoral Research Fellow at the Center for International Development at the Harvard Kennedy School. She is founder and co-chair of the Oxford Summer School in Economic Networks, a bi-annual multi-disciplinary school for over 100 postgraduate students. She holds a PhD (mathematics) from Imperial College, and was founder and Editor in Chief of Angle – a journal based at Imperial College focusing on the intersection of policy, politics and science – between 2009-2020.

Adaptor signatures are an extension of standard digital signatures that tie together the creation of a digital signature (e.g., authorising a blockchain transaction) and the leakage of a secret value (e.g., the pre-image of a hash value as in hash-time lock contracts). In a nutshell, one can first generate a pre-signature with respect to a secret value, which can be converted to a valid signature only by knowing the secret. Second, if the pre-signature is converted to a valid signature, one can extract the secret from the pair (pre-signature, valid signature). In this talk, we will overview the notion of adaptor signatures and how these two properties can be used as building block for several blockchain applications (e.g., atomic coin/token swaps or multi-hop payments) that provide strong privacy and scalability guarantees.

A nonequilibrium system is characterized by non-vanishing thermodynamic forces and fluxes, which give rise to entropy production (EP). We demonstrate that these forces and fluxes have an information-geometric structure, which allows EP to be decomposed into nonnegative contributions from different types of forces. We focus on the excess and housekeeping decomposition, which reflects contributions from conservative and nonconservative forces, in the general setting of discrete systems (linear master equations and nonlinear chemical dynamics). Unlike the nonadiabatic/adiabatic (Hatano-Sasa) approach, our decomposition is always well-defined, including in systems with odd variables and nonlinear systems without steady states. Our decomposition is operationally meaningful, leading to far-from-equilibrium thermodynamic uncertainty relations and speed limits.

Agent-based models (ABMs) are deterministic-stochastic maps that iterate the state of a system forward in time. In standard practice, ABM microstates -representing heterogeneous, interacting units- are initialized randomly and evolved following internal dynamics that are not anchored to real-world time series. In recent years, however, more and more researchers have started to initialize their ABM microstates with real-world data and attempted to reproduce real-world dynamics. In this talk, I will give an overview of my research on data-driven economic ABMs. I will discuss theoretical problems such as latent variable estimation and consistency between micro and macro economic statistics, and show two applications. In the first application, I will address some of the most debated issues related to epidemic-economic tradeoffs by introducing a detailed ABM that simulates infections and unemployment at the level of 500,000 synthetic individuals in the New York area. This ABM is initialized from census data, regional and national accounts and input-output tables, and cell phone mobility data. In the second application, I will discuss the effect of beliefs about sea level rise on the housing market, by building an ABM of the housing market of Miami that uses detailed property, transaction, mortgage and demographic data. Overall, these results suggest that ABMs are the ideal tool to bridge big data and theoretical modeling, although a lot of research is still needed to develop standard techniques and practices.

Economic Complexity (EC) algorithms estimate the fitness and complexity of the nodes of bipartite networks. Typical examples are the network of countries and their exported products or the network of countries and the fields of scientific production. These algorithms cease to work as soon as the network is not bipartite anymore, even if there is only one “weak” link between two nodes of the same class. Our task has been to generalize EC algorithms to deal with non-bipartite networks. We first analyze the linear Economic Complexity Index (ECI) method and then the non-linear Economic Fitness Complexity (EFC), showing that the latter is more stable after introducing small non-bipartite perturbations. Eventually, we use EFC to study the complexity of the prey-predator ecosystem in Florida Bay. We recommend this seminar to all our colleagues who deal with those systems where it is essential to estimate the importance of the actors involved, not necessarily systems in economics.

Learning with deep neural networks has enjoyed huge empirical success in recent years across a wide variety of tasks. Despite being a complex, non-convex optimization problem, simple methods such as stochastic gradient descent (SGD) are able to recover good solutions that minimize the training error. More surprisingly, the networks learned this way exhibit good generalization behavior. Understanding generalization is one of the fundamental unsolved problems in deep learning. This problem has been studied extensively in machine learning, with a rich history going back more than 50 years. However, most of existing theories in machine learning fail when applied to modern deep networks. In this talk we will have a practical look on neural networks through the lens of in-distribution and out-of-distribution generalization.

The Corona data donation project (Corona-Datenspende) is one of the largest citizen science projects worldwide in which over 100,000 people donate personalized, physiological sensor data (heart rate, physical activity and sleep) collected via fitness trackers and smart watches in a privacy preserving fashion. At the same time users take part in regular surveys about their current and past health status as well as their behaviour and experiences over the course of the COVID-19 pandemic. The scientific progress is made transparent on a regular basis via social media and a dedicated science blog that instigates discussion and interaction among the users and the involved researchers. Here, we showcase a variety of applications and insights that have evolved from this unique data set: (i) A nowcast system for COVID-19 case numbers from physiological data. (ii) A classification system for Long-Covid that shows the positive effect of vaccinations for mitigating long-term impacts on individual health. (iii) A large-scale analysis of human sleep patterns that shows pronounced geographical and socio-economic imprints. (iv) The overall positive relationship between physical activity, sleep and self-assessed well-being. Our results imply that digital projects like the Corona-Datenspende have the potential for engaging a large portion of the general public in addressing the contemporary challenges of our time. Due to its flexibility it simultaneously complements traditional methods such as field-work, clinical studies or surveys as data collection can be quickly adapted to new emerging research questions.

I will be covering some modelling issues that come up when considering the long-term development of societies. Of particular importance is the topic of societal collapse as the archaeological record has numerous instances of the phenomenon. I will discuss some of the general modelling philosophy, relevant literature, my own work on ancient societies (Easter Island, the Maya, Roman Empire and Chinese dynasties) and implications for modern society. There are several modelling considerations unique to modern society that will be highlighted.

In the era of globalisation, supply chains are increasingly international, thus disconnecting the location of production from final consumption. Consumption has developed into a major, geographically distant driver of various local environmental impacts in countries producing raw materials. Despite continuous developments, the spatial resolution of methods to assess global supply chains from raw material extraction to final demand and to calculate consumption-based (or footprint) indicators has been limited to the national level. This leads to distorted results, as the heterogeneity of environmental conditions within producing countries is not taken into account. In my talk, I will introduce novel assessment frameworks developed in the ERC FINEPRINT (www.fineprint.global) and other projects that allow quantifying material footprints and related environmental impacts on a high spatial detail. The framework includes assessments of the geographical distribution of raw material extraction in countries world-wide and linking these global extraction maps to spatially explicit data on environmental impacts, to address issues such as land use change, deforestation or water scarcity. We also develop multi-regional input-output models that include sub-national information in major resource extraction countries, in order to trace raw material flows and related impacts along global supply chains to the country of final demand. These novel approaches improve the understanding of the relations between global drivers and local impacts in hot-spot extraction regions and supply chains. Our results are relevant to a wide range of policy initiatives to mitigate the environmental impacts of resource extraction and to achieve more sustainable production and consumption patterns.

The foundation of modern societies is institutions like governments, corporations, and even emergent social communities. We might think of institutions as singular entities, but they are composed of multitudes of disparate people, documents, and traditions. While we have an intuitive sense of what institutions are, they are hard to define. To remedy that, we aim to ground our intuition on a well-documented and relatively stable model system, the United States Courts of Appeals. Importantly, the system consists of an ensemble of several nominally independent courts, which permits a comparative study. I will focus on a central aspect of this institution, the judges’ written opinions, and explore ways of representing them in a meaningful way. I use text embedding techniques to map the textual features into descriptive document vectors. I characterize the results of popular methods like Doc2Vec and Legal-BERT. I will show how we evaluate the embeddings in terms of their predictive ability, alignment with the citation structure, and qualitative sensibility.

The main goal of this project is to understand the effects of declining expectations about future economic growth rates on financial instability. As a first step, a heterogenous agent model of a stylised stock market is analysed. Such models are useful tools to model endogenous fluctuations in financial markets, caused by self-reinforcing dynamics of trend-following trading strategies. In the presented work, the focus is on the reaction of such model dynamics to an external shock, caused by the decline of expected future dividend growth rates. The dynamic is modelled with two representative trader types, fundamentalists and chartists, as well as evolutionary switching behaviour between the two strategies. Preliminary results of stability analysis and numerical simulations are presented.

This research project presents a complex, modular, 1:1 scale model of the Hungarian residential housing market. All the 4 million households and their relevant characteristics are represented based on empirical micro-level data coming from the Central Credit Information System, the Pension Payment database and transaction data of property sales collected by the National Tax and Customs Administration and the largest real estate agencies. The model features transactions in the housing and rental markets, a construction sector, buy-to-let investors, housing loans, house price dynamics and a procyclical banking sector regulated by a macroprudential authority. The flats in the model are characterized with detailed attributes regarding their size, state and neighbourhood quality. Households choose the flat with the highest consumer surplus according to standard utility maximization theory. Additionally, we have also implemented demographic trends, including childbearing, marriage and inheritance. This way the model is suitable for analysing various types of macropudential, fiscal and monetary policies as well as for the assessment of exogenous shock scenarios. Initiating the model simulation from 2018, it managed to reproduce the number of transactions and the observed house price dynamics in most of the regions of Hungary for 2018-2019, while the volume of new housing loans and their distribution regarding income deciles and loan-to-value ratios were also in compliance with the empirical data.

tba

The vulnerability of supply chains and their role in the propagation of shocks has been highlighted multiple times in recent years, including by the recent pandemic. However, while the importance of micro data is increasingly recognised, data at the firm-to-firm level remains scarcely available. We formulate the reconstruction of supply chain networks as a link prediction problem and tackle it using machine learning. We test our approach on three different supply chain datasets and show that it works very well and outperforms three benchmarks. An analysis of features’ importance suggests that the key data underlying our predictions are firms’ industry, location, and size. Using different datasets for training and testing, we evaluate our approach’s effectiveness in a real-world scenario when no production network data is available.

Right-wing authoritarianism (RWA) is on the rise, but authoritarians’ daily behaviors remain uncharted. RWA research has so far relied predominantly on data from self-reports and there are only few findings on objective indicators of RWA. Here, digital traces from smartphone use represent a promising means to investigate the behaviors of new authoritarians as they go about their modern lives. To this end, we drew on RWA literature to derive a comprehensive overview of theoretical statements and empirical reports on behavioral indicators of authoritarianism. We then translated these findings into behavioral features which can be captured using data collected from smartphone sensors and logs (e.g., app-use, mobility, music/podcast consumption). Lastly, we use machine learning models to predict self-reported authoritarianism from these behavioral features, using data from a representative sample of 749 participants whose smartphone use was tracked continuously for up to six months. By creating a theory-informed profile of authoritarians in the digital era, we aim to contribute to the containment of the spread of authoritarianism. Bio: Alejandro is a PhD student at the LMU Munich School of Management working at the intersection of organizational behavior, social psychology, and data science. He holds a BSc in Psychology from the UNAM (Mexico) and an MSc in Economic, Social, and Organizational Psychology from the LMU Munich. Alejandro is interested in the study of the self and identity generally, and multiple sources of identity (e.g., family, work, political affiliation) specifically. He uses data from online communities, smartphones, and surveys to examine pressing contemporary phenomena such as mandatory work from home, job loss, and authoritarianism.

Using digital traces from online dating gives us the opportunity to study the earliest stages of human mating. We focus on whether online dating app users are homophilic in terms of the desirability of whom they pursue. Using data from a Czech online dating app, we construct networks where nodes represent users and ties represent messages expressing interest (“swipes”). We find that the structure of the networks is considerably hierarchical, with women having the upper hand on the app as they are in the “choosing position” due to the uneven gender ratio on the app and their substantially higher desirability. The results further show that individuals initially pursue users who are more desirable than themselves. The reciprocated contacts are comparatively more homophilic. These results suggest that in terms of desirability, the similarity of partners is due to the subsequent mating processes (e.g., rejection) rather than due to initial preference for similarity. Bio: Renáta is a PhD student at the department of Sociology at Charles University and a researcher at the department of Social Stratification at the Institute of Sociology of the Czech Academy of Sciences. Renáta’s main research interests lay in the application of computational social science and the study of various sociological phenomena, such as social stratification, scientific communication, and most prominently online dating markets. To tackle these issues, she employs a wide range of methods from quantitative text analysis, online experiments to social network analysis.

The social and ecological challenges of our time require a better understanding of the complex interactions between the multiple dimensions of human well-being and environmental impacts. This work introduces the Needs and Limits (N&L) framework, a theoretical and computational foundation for agent-based simulations of human individuals who try to increase their quality of life through the satisfaction of human needs. Based on psychological research, human needs are described as heterogeneous, satiable, adaptive, and interdependent with the social and bio-physical environment. The N&L framework represents a generic foundation that can be applied to a broad range of socio-economic and ecological scenarios, which is illustrated for the topics of income inequality and climate policy.

1. The Covid-19 pandemic drastically emphasized the fragility of national and international supply networks (SNs), leading to significant supply shortages of essential goods for people, such as food and medical equipment. Severe disruptions that propagate along complex SNs can expose the population of entire regions or even countries to these risks. A lack of both, data and quantitative methodology, has hitherto hindered us to empirically quantify the vulnerability of the population to disruptions. Here we develop a data-driven simulation methodology to locally quantify actual supply losses for the population that result from the cascading of supply disruptions. We demonstrate the method on a large food SN of a European country including ~23.000 business premises, ~44,000 supply links and 116 local administrative districts. We rank the business premises with respect to their criticality for the districts’ population with the proposed systemic risk index, SRIcrit, to identify around 30 premises that—in case of their failure—are expected to cause critical supply shortages in sizable fractions of the population. The new methodology is immediately policy relevant as a fact-driven and generalizable crisis management tool. This work represents a starting point for quantitatively studying SN disruptions focused on the well-being of the population. - 2. Did you ever: swear at your screen for hours because of an unknown bug in your code? Wait an infinite time in front of the computer just to get an incomplete result? Spot inconsistency in your source data really late in your project? Forget until the day of the deadline about a random assumption that was made at the beginning? Argue with a collaborator about which version of the code was used for generating data_final_2.csv? Using concepts from software engineering good practices, you can forget all these lower significantly their frequency. I will go through some of the software design choices that we made.

Johannes will introduce three new additions to our VisTool and PyVisTool. These will let you upload and host your own data in order to explore: – geospatially distributed data, and correlations between them – geospatial flows, i.e., origin-destination data in a geographical context – networks in 3D (and 2D) There will also be a tutorial on how to access / interface with the VisTool and PyVisTool.

We consider a generalised d-dimensional model for asymptotically-scale-free geographical networks. Central to many networks of this kind, when considering their growth in time, is the attachment rule, i.e. the probability that a new node is attached to one (or more) preexistent nodes. In order to be more realistic, a fitness parameter ?_i for each node i of the network is also taken into account to reflect the ability of the nodes to attract new ones. Our d-dimensional model takes into account the geographical distances between nodes, with different probability distribution for ? which sensibly modifies the growth dynamics. The preferential attachment rule is assumed to be ?_i?k_i*?_i*r^(??_A) where k_i is the connectivity of the i–th pre-existing site and ?_A characterizes the importance of the euclidean distance r for the network growth. For special values of the parameters, this model recovers respectively the Bianconi–Barabási and the Barabási–Albert ones. The present generalised model is asymptotically scale-free in all cases, and its degree distribution is very well fitted with q-exponential distributions, which optimizes the nonadditive entropy Sq, given by p(k)?e^(?k/?)_q?1/[1+(q?1)k/?]^(1/(q?1)), with (q,?) depending uniquely only on the ratio ?_A/d and the fitness distribution. Hence this model constitutes a realization of asymptotically-scale-free geographical networks within nonextensive statistical mechanics, where k plays the role of energy and ? plays the role of temperature. General scaling laws are also found for q as a function of the parameters of the model.

The transition to a low carbon economy comes with many challenges. One of the biggest is the reorganization of economic production such that the least amount of greenhouse gases is emitted while the production of economic goods and services is kept at decent levels. We model the economy as a complex network to show how a networks perspective might inform the low carbon transition. We compare the systemic relevance of each company with its greenhouse gas emissions to identify potential levers for decarbonization. This idea is demonstrated for the 30.000 companies involved in the Austrian pork supply network which we reconstructed from various data sources. The nodes in this network are individual companies such as farms, slaughterhouses, meat processors, distribution centers and supermarkets. The edges are supply and buy relations between those companies and represent the number of transferred pigs or pork. Each company is dependent on its suppliers to deliver resources for its production and on its buyers to sell its products. From these interdependencies network structures arise that are vulnerable to potential cascading failures, should a node stop supplying and/or buying its neighboring nodes. The inherent systemic risk that each node poses to the whole network can be quantified by simulating its failure and observing the consequences. By summing up the losses of production one can calculate a company’s economic systemic risk index (ESRI) – a method developed by Diem et al. [1]. We compare this measure of socio-economic relevance with the CO2 emissions of each company in the Austrian pork supply network. We know the CO2 emissions associated with a particular production step from a recent lifecycle assessment of Austrian pork done by Winkler et al. [2]. These emission factors are used to calculate an estimate of each company’s annual CO2 emissions. We find that companies with high emissions and low socio-economic relevance are potential levers for decarbonization. By targeting these companies with regulatory policies or tailored carbon taxes a maximum of saved emissions and a minimum of disruption to overall production can be expected. Modeling the economy as a complex network therefore reveals mitigation potentials and potential pitfalls. We plan to extend this proof of concept to more complex supply networks in the future. References: [1] C. Diem, A. Borsos, T. Reisch, J. Kertesz, S. Thurner, arXiv:2104.07260v1 [2] T. Winkler, K. Schopf, R. Aschemann, W. Winiwarter, J Cleaner Production, 116 (2016) 80-89

Seshat aims to build the most comprehensive body of knowledge about human history in one place. In this talk, I will explain the workflow involved in using the Seshat Databank, from a Data Science perspective. I will also talk about the main concepts of database management, data cleaning, and how we extract knowledge from Seshat Databank. In order to make Seshat more available to all interested individuals around the world, we plan to build more user-friendly interfaces, where Seshat RAs, Seshat experts, and even the general public can communicate with the data efficiently. In this regard, I will talk about the basics of data models, SQL, Jupyter Notebook, Django, etc. and how we are using these tools in Seshat publications as well as our future database and website. One of our goals is to go beyond traditional database management systems towards graph databases, and in that front, we are in close collaboration with TerminusDB. As this is a skillup webinar, I will also talk about some other tools that I find interesting, such as Trello and Notion.

Explicitly “moralizing” religions such as Christianity or Buddhism — those which posit a system of supernatural punishment and reward for interpersonal ethical conduct — emerged relatively late in human history. Yet they have now spread to all corners of the globe. Why did they arise and how have they become so successful? In this brief talk I’ll introduce the Seshat History of Moralizing Religion, a multi-author edited volume being co-edited by Jennifer Larson (a historian of ancient Greek religion), Peter Turchin (an evolutionary anthropologist based at the CSH), and myself. The book leverages Seshat: The Global History Databank’s sampling framework, deep historical coverage and close connections with historians and archaeologists to offer the first worldwide comparative perspective on moralizing religions.

We mine the leaked history of trades on Mt. Gox, the dominant Bitcoin exchange from 2011 to early 2014, to detect the triangular arbitrage activity conducted within the platform. The availability of user identifiers per trade allows us to focus on the historical record of 440 investors, detected as arbitrageurs, and consequently to describe their trading behavior. We begin by showing that a considerable difference appears between arbitrageurs when indicators of their expertise are taken into account. In particular, we distinguish between those who conducted arbitrage in a single or in multiple markets: using this element as a proxy for trade ability, we find that arbitrage actions performed by expert users are on average non-profitable when transaction costs are accounted for, while skilled investors conduct arbitrage at a positive and statistically significant premium. Next, we show that specific trading strategies, such as splitting orders or conducting arbitrage non aggressively, are further indicators of expertise that increase the profitability of arbitrage. Most importantly, we exploit within-user (across hours and markets) variation and document that expert users make profits on arbitrage by reacting quickly to plausible exogenous variations on the official exchange rates. We present further evidence that such differences are chiefly due to a better ability of the latter in incorporating information, both on the transactions costs and on the exchange rates volatility, eventually resulting in a better timing choice at small time scale intervals. Our results support the hypothesis that arbitrageurs are few and sophisticated users.

Spreading processes play an increasingly important role in modeling for diffusion networks, information propagation, marketing and opinion setting. We address the problem of learning of a spreading model – including the spreading network structure – such that the predictions generated from this model are accurate and could be subsequently used for the optimization, and control of diffusion dynamics. We focus on a challenging setting where full observations of the dynamics are not available or noisy, and standard approaches such as maximum likelihood quickly become intractable for large network instances. We introduce a computationally efficient algorithm, based on a scalable dynamic message-passing approach, which is able to learn both parameters of the effective spreading model and network structure, given only limited information on the activation times of nodes in the network. The popular Independent Cascade model is used to illustrate our approach. We show that tractable inference from the learned model generates a better prediction of marginal probabilities compared to the original model. We develop a systematic procedure for learning a mixture of models, which further improves the prediction quality. Bio: Mateusz obtained his PhD in Physics from the University of Warsaw. He later worked as a Postdoc in the Quantitative Finance Research Group of Scuola Normale Superiore di Pisa. Currently, he is a Postdoc in Plasma Physics and Applied Mathematics Group at the Los Alamos National Laboratory. He has broad scientific interests, which include subjects from statistical physics, network science and machine learning. More specifically, he is interested in belief propagation techniques for network problems, epidemic spreading modelling, financial networks etc.

Human decision-making in social situations is affected by a diversity of factors including material cost-benefit considerations, normative and cultural influences, learning, and conformity with peers and external authorities (e.g., cultural, religious, political, organizational). Also important are their dynamically changing personal perception of the situation and beliefs about actions and expectations of others as well as psychological phenomena such as cognitive dissonance and social projection. Recently I developed a unifying mathematical framework capturing a diversity of the above factors (Evolutionary Human Sciences 3: e44 (2021)). In this talk, I’ll discuss the results of a long-term experimental economics study testing, validating, and parameterizing my general model within the context of the Common Pool Resources game. Our approach allows us to evaluate and compare the importance of different material and non-material factors in both statistical and economically meaningful ways.

Research has repeatedly shown that exposure to suicide-related news media content is associated with suicide rates, with some content characteristics likely having harmful and others potentially protective effects. Although good evidence exists for a few selected characteristics, like reporting on celebrity deaths by suicide, systematic and large scale investigations of many other characteristics are missing. Moreover, the growing importance of social media, particularly among young adults, calls for studies on the effects of content posted on these platforms. This study used natural language processing and machine learning methods to automatically label large quantities of social media data according to characteristics considered important for media effects research on suicide. We manually labelled 3200 English tweets using a novel annotation scheme, which differentiates postings based on their type of topic, underlying problem- vs. solution-focused narrative, and serious vs. nonserious/metaphorical use of suicide-related terms. After splitting this dataset into a training and test set, we trained different machine learning models, including a more traditional (TF-IDF) as well as two state-of-the-art deep learning models (BERT, XLNET), on several classification tasks. Most importantly, we classified postings into six content categories that might differentially affect suicidal behavior: personal stories of either suicidality or coping (i.e., Papageno-related tweets), general messages intending to spread either awareness or prevention-related information, reportings of suicide cases (i.e., Werther-related tweets), and other suicide-related or off-topic tweets. In a further task, we separated postings that refer to actual suicide from those that use suicide-related terms in a metaphoric, sarcastic or other irrelevant way. In both tasks, the performance of the deep learning models was similar, and much better than the traditional approach. When classifying the six content types, the BERT model correctly classified 74% of tweets in the test set, and F1-scores lay between 55% to 85% for the different categories of interest (above 70% for all but the suicidality category). Furthermore, BERT correctly labelled 88.5% of tweets about vs. not about suicide in the test set, achieving F1-scores of 92% and 73% for the two categories. These classification performances are comparable to the state-of-the-art on similar tasks, and demonstrate the potential of machine learning for media effects research. By making data labelling more efficient, this work will enable future large-scale investigations on harmful and protective effects on suicide rates and help seeking behavior for different characteristics of suicide-related content.

The variance of a probability distribution is a fundamental concept in the toolkit of probability theory and statistics and is routinely applied throughout science, engineering and numerous settings. In many practical cases however, probability distributions are defined on the nodes of a network. As the usual definition of variance cannot incorporate the network topology into the computation, we lack a basic methodological tool when analyzing distributions on networks. To bridge this methodological gap, we have developed a theory to measure the variance and covariance of probability distributions defined on the nodes of a network, which incorporates the topology of the network by considering the distance between nodes. Our approach generalizes the usual (co)variance to the setting of weighted networks and retains many of its intuitive and desired properties. To illustrate the application of these new measures in practice, we use them to analyze two empirical networks of mathematical concepts built with data from Wikipedia and a collection of scientific papers retrieved from ArXiv. Our approach allows for a unified and intuitive treatment of the structural data (relations between concepts) and functional data (usage of concepts in papers). Since the variance and covariance are general-purpose statistical tools, these new metrics may find application in multiple fields, like neuroscience, economics or social network analysis.

* New Vis Landing Page: A central place where visualizations, apps and other visual scientific output can be hosted and accessed. * Hub Vis: explore the network of CSH researchers and affiliates based on their scientific collaborations (and more). * Vis Tool & PyVis Tool: an interface to upload and explore your network data directly through our in-house created visualizations.

Why do some social groups or organizations, initially composed of a few people with the same interests, disintegrate when increasing the number of members? To answer this question, we propose a stochastic model where group members and their relationships are described by a growing signed network with positive and negative signs on both nodes and links. We investigate, both analytically and numerically, the behavior of group cohesion, defined as the fraction of nodes with the same sign, when the network grows with noisy information on the link signs. Our findings suggest strategies to prevent or boost group cohesion in different contexts.

The gender pay gap in the United States has been closing since the 1980s, the same period in which the American economy computerized. We study whether this closing happened despite computerization, or because of it. To answer this, we study whether the closing of the pay gap mainly happened because of gender differences in the degree of labor reallocation between occupations, or because of gender differences in the transformation of the job tasks inside occupations. The latter would suggest a more straightforward technological explanation. We also analyze the pre-computer technologies and job tasks in male and female jobs, and study whether these initial conditions contributed to the faster de-routinization of female-dominated jobs that we observe. We construct a new longitudinal dataset that makes such analysis possible. Using optical character recognition and natural language processing, we transformed the U.S. Dictionary of Occupational Titles (DOT, 1939 – 1991) into a database akin to, and comparable with its digital successor, O*NET (1998 – today). After creating a single occupational classification, we connected all DOT waves and decennial O*NET databases into a single dataset stretching over nine decades, and merged into this information from the U.S. Decennial Census on employment, wages, education, and other demographic and labor market characteristics. This is joint work with Shreyas Gadgin Matha, James McNerney, Andre Assumpcao, Dario Diodato, and Frank Neffke.

The net zero transition involves a shift in demand from emission-intensive industries, such as oil and coal, to green industries. In this lecture, we survey what the literature has found about the implications of the green transitions for job creation and destruction of jobs. We then discuss the role skills and occupational mobility may play in employment, and how networks and agent based models can help us better prepare workers for the green transition. Finally, we conclude with further work and the possibility of building a data-driven agent based model linking both the production sector and the labor market sector for the green transition.

Media reporting on suicide has repeatedly been shown to be associated with suicide rates. The impact of suicide reporting may not be restricted to harmful effects; rather, stories of coping and recovery in adverse circumstances may have protective effects. Specifically, exposure to media reports about deaths is associated with increases in suicides, suggesting a Werther effect. In contrast, exposure to content describing stories of hope and coping are associated with a decrease in suicides, which has been labeled as the Papageno effect. Investigating the impacts of suicide reporting requires classifying various characteristics of media-items that may have harmful or negative effects, which proves time-intensive and challenging. Using natural language processing for the classification of such texts could facilitate this tedious task. We use the bidirectional language model BERT and compare its performance against TFIDF and Bag-of-words. We show that deep learning and synthetic data generation allow developing an application, which is capable of processing English texts and detecting specific characteristics of suicide-related content. We describe an effective classification model that enables the user to retrieve the predicted label of a specific variable code for the given English input text. Simple binary classification tasks are best solved by a fine-tuned BERT model trained on the original data, achieving 85%?95%F1 compared to human performance of F1human=100%. Intermediate binary classification tasks often benefit from synthetically balancing the data, with performances around 75%-80% (F1human?80%).. Difficult binary classification and multi-class classification tasks always benefit from synthetically balancing the data. However, which balancing method works best is task specific, and performances range between ?70%(F1human?80%)and?80%(F1human?95%), respectively. Our results show that pre-trained bidirectional language models work incredibly well. Yet, improvements seem to mostly come from bigger models and more data. Synthetically balancing the minority classes provides more training data and improves the model’s ability to generalize to new inputs. However, limiting the amount of synthetic data is crucial, since performance appears to tail off when the balance is tipped too far in favour of the synthetic data. Our application will enable researchers to investigate the effect of different characteristics of texts about suicide at large scales and help improve reporting guidelines, thereby effectively contributing to the prevention of suicides.

The history of a university is typically told through the annals of its professors, deans and rectors. To offer an additional perspective, we study the development of medical research in Vienna from 1930 to 2020 within the framework of quantitative history. To this end we perform an analysis of the collaborations, publication activity and international reception of medical research conducted at the University of Vienna and associated institutions. Based on the database Dimensions.ai we use the published affiliations of researchers to construct the network of international collaborations for Viennese medical research. Furthermore, we analyse the amount of produced research by publications and investigate the citation networks with respect to the influence of Viennese works on international publications as well as international influences on Viennese researchers. To compare our indicators with leading international institutions, we pick Harvard Medical School, the Karolinska Institute and Utrecht Medical School and discuss temporal and regional differences. This allows us to identify general trends as well as local characteristics. We find changes in the intra- and inter-university cooperation networks, most prominently their strong densification since 2000. During the 1930s and 1940s national politics and World War II led to international isolation of Viennese research institutions and a dramatic reduction in publishing activity, transforming Vienna from a world-leading location for medical science to marginality. We discuss factors that help or hinder the catching-up process to world-leading places of research that is still continuing to date. The discussed factors include the transition towards English as the main language of publication, more international connections in the citation network and changes in the intra- and inter-institutional collaboration network. We discuss the trends in medical research and regional specializations by analysing the composition of a university’s publications according to fields of research. The affiliations used in scientific publications reveal the mobility of researchers and allow us to discuss the in- and outgoing mobility of authors for the investigated institutions. We conclude with remarks on the limitations of our study and historical bibliometric data in general. With our case study we want to highlight the potential of scientometrics for historical analyses.

Data sharing is crucial for building large data sets in different fields of application. Data privacy is a legitimate concern of subjects contributing to the data. Even when all unique identifiers like names or identification numbers are removed from a dataset, the dataset is not anonymized but rather pseudonymization. It is still possible to identify individuals in a given pseudonymized dataset, when linking pseudonymized data with publically available data. This is especially true for time series sensor data, which consist of a sequence of data points rather than just a single data record. In this talk, I am going to show how to localize the origin of sensor data by combining it with publically available weather data. With our method we are able to locate cows with an average localization accuracy of 28.2 km over the area of Austria with nothing more than a trace of sensor data measured in their rumen. For sensor readings from O3 data and solar data, we are even able to lower the average error to 25.9 km and 7.4 km respectively. I will illustrate how to perturb the data to avoid this kind of privacy leakage. I will also illustrate how to tune the privacy-utility trade-off.

Optimizing a neural network is often investigated as finding a minimum in an objective landscape. Therefore, understanding the geometric properties of this landscape has emerged as an important goal. Related works show that independently trained models are connected by a curve in weight space along which loss remains low. This curve could be a line if these trained models are residing in the same basin. We know many different hyperparameters affect where the endpoints stand, e.g. networks that share only a few epochs of their optimization trajectory are converging in the same basin, hence connected by a linear path of high accuracy. Aligned with this literature, in this talk, we try to understand how we can predict the geometry of the loss landscape. We also conjecture that if the permutation invariance of neural networks is taken into account, SGD solutions will likely have no barrier in the linear interpolation between them.

Background: Young refugees are at increased risk of labour market marginalisation (LMM). We examined whether the association of multimorbidity patterns and LMM differs in refugee youth compared to Swedish-born youth, and identify the diagnostic groups driving this association. Methodology: We analyzed 249,245 individuals between 20-25 years at 31.12.2011 from a combined Swedish registry. Refugees were matched 1:5 to Swedish-born youth. A multimorbidity score was computed from a network of disease co-occurrences in 2009 – 2011. LMM was defined as disability pension (DP) or > 180 days of unemployment during 2012 – 2016. Relative risks (RR) of LMM were calculated for 114 diagnostic groups (2009 – 2011). The odds of LMM as a function of multimorbidity score were estimated using logistic regression. Results: 2841 (1.1%) individuals received DP and 16,323 (6.5%) experienced >180 annual days of unemployment during follow-up. Refugee youth had a marginally higher risk of DP (OR(95%CI): 1.59(1.52, 1.67)) depending on their multimorbidity score compared to Swedish-born youth (OR(95%CI): 1.51 (1.48, 1.54)); no differences were found for unemployment (OR(95%CI): 1.15 (1.12, 1.17), 1.12 (1.10,1.14), respectively). Diabetes mellitus and influenza/pneumonia elevated RR of DP in refugees (RRs (95% CI) 2.4 (1.02, 5.6) and 1.75 (0.88, 3.45), respectively); the majority of diagnostic groups were associated with a higher risk for unemployment in refugees. Conclusion: Multimorbidity related similarly to LMM in refugees and Swedish-born youth but different diagnoses drove these associations. Targeted prevention, screening, and early intervention strategies towards specific diagnoses may effectively reduce LMM in young adult refugees.

With ever wider aspects of the human and non-human world being datafied, and research in medicine and many other areas becoming increasingly datafied, the field of data ethics is becoming more important. But what is it? What is the difference between data ethics and other fields of ethics, such as medical ethics? And does it create more problem than it solves? These and other questions will be at the centre of our seminar.

The global scientific output in terms of numbers of publications has been increasing rapidly—exponentially—in recent years, replacing and outdating historical knowledge at a tremendous rate. Only a few scientific milestones of exceptional quality have remained relevant over decades and consistently attract citations. From the perspective of scientometrics and science funding, this raises important questions. For how long is scientific work relevant? How long will it take before today’s work is forgotten? And how are milestone papers remembered differently? To answer these questions, we study the temporal citation network of Physics works published in American Physical Society journals. We present a model to quantify the probability of attracting citations for individual publications based on their age and the number of citations they received in the past. We identify the bulk of publications that is forgotten in a largely homogeneous way as well as exceptional works that remain relevant for long periods of times. We capture the forgetting as well as the tendency to cite popular works in a microscopic generative model of scientific citation networks. The model allows us to understand how some papers are cited while others are forgotten in the context of a growing citation network. Moreover, it allows us to estimate an expected citation landscape of the future, predicting that 95% of papers relevant in 2050 have not yet been written. We identify some conditions under which individual publications manage to achieve immortality.

During the current pandemic we have seen a multitude of non-pharmaceutical mitigation measures, which may largely be grouped into endogenous self-distancing/social avoidance and exogenous isolation measures. Here we are studying a minimal compartmental model that aims to understand the relative effect of self-distancing and quarantining. We introduce a simple adaptive extension of the SIR model with a quarantine compartment. The latter is known as the SIRX model, but to our knowledge the adaptive extensions are novel. First, we compare computationally expensive, adaptive network simulations with their corresponding computationally ODE equivalents and find excellent agreement. Second, we discover that there exists a relatively simple critical curve in parameter space for the epidemic threshold, which strongly suggests a mutual compensation effect between the two mitigation strategies: as long as social distancing and quarantine measures are both sufficiently strong, large outbreaks are prevented. Third, we study the total number of infected and the maximum peak during large outbreaks using a combination of analytical estimates and numerical simulations. Also for large outbreaks we find a similar compensation effect as for the epidemic threshold. This suggests that if there is little incentive for social distancing within a population, drastic quarantining is required, and vice versa. Both pure scenarios are unrealistic in practice. Our models show that only a combination of measures is likely to succeed to control epidemic spreading. Fourth, we analytically compute an upper bound for the total number of infected on adaptive networks, using integral estimates in combination with the moment-closure approximation on the level of an observable.

Previous work utilizing Seshat: Global History Databank, a major resource for studying patterns of sociocultural evolution in world history, has revealed that complex societies from across the globe and at different time periods develop, spread, and collapse in a largely consistent manner, incorporating a package of ‘technologies’ from productivity-enhancing implements like iron plows to forms of governance to normative ideological systems. We theorize that much of these developments are driven by intense inter-state competition and the evolutionary demands of – and limits to – growing complexity. Here, I will discuss these findings and highlight our recent work trying to uncover the main causal forces behind the development of military technologies, a key factor in and indicator of intense interstate competition. Understanding the evolution of critical military technologies is a complex but important project, as delineating the causes and consequences of the adoption of various military technologies can help us understand not only the evolution of technology generally, but also carries far-reaching implications for the dynamics of social complexity more broadly. I conclude by previewing current efforts within the Seshat project linking these insights on interstate conflict to the dynamics of internal competition and how this research can help shed light on the rise and fall of states, in the past and, perhaps, in the modern world as well. About Dr. Dan Hoyer: Daniel Hoyer is Project Manager of Seshat: Global History Databank and Part-Time Professor at the George Brown College Centre for Preparatory and Liberal Studies. He holds a PhD in Classics from New York University, where he studied economic and social development in the high Roman Empire. His current research employs comparative historical and social scientific methods to explore the causes and limiting factors to economic growth, societal development, and general well-being. In particular, he is interested in understanding the role of prosocial cultural traits in promoting equitable distribution of resources and limiting predatory activity in past societies.

Crises like COVID-19 or the Japanese earthquake in 2011 exposed the fragility of corporate supply networks. The production of goods and services is a highly interdependent process and can be severely impacted by the default of critical suppliers or customers. While knowing the impact of individual companies on national economies is a prerequisite for efficient risk management, the quantitative assessment of the involved economic systemic risks (ESR) is hitherto practically non-existent, mainly because of a lack of fine-grained data in combination with coherent methods. Based on a unique value added tax dataset we derive the detailed production network of an entire country and present a novel approach for computing the ESR of all individual firms. We demonstrate that a tiny fraction (0.035\%) of companies has extraordinarily high systemic risk impacting about 23\% of the national economic production should any of them default. Firm size alone cannot explain the ESR of individual companies; their position in the production networks does matter substantially. If companies are ranked according to their economic systemic risk index (ESRI), firms with a rank above a characteristic value have very similar ESRI values, while for the rest the rank distribution of ESRI decays slowly as a power-law; 99.8\% of all companies have an impact on less than 1\% of the economy. We show that the assessment of ESR is impossible with aggregate data as used in traditional Input-Output Economics. We discuss how simple policies of introducing supply chain redundancies can reduce ESR of some extremely risky companies.

The global scientific output in terms of numbers of publications has been increasing rapidly—exponentially—in recent years, replacing and outdating historical knowledge at a tremendous rate. Only a few scientific milestones of exceptional quality have remained relevant over decades and consistently attract citations. From the perspective of scientometrics and science funding, this raises important questions. For how long is scientific work relevant? How long will it take before today’s work is forgotten? And how are milestone papers remembered differently? To answer these questions, we study the temporal citation network of Physics works published in American Physical Society journals. We present a model to quantify the probability of attracting citations for individual publications based on their age and the number of citations they received in the past. We identify the bulk of publications that is forgotten in a largely homogeneous way as well as exceptional works that remain relevant for long periods of times. We capture the forgetting as well as the tendency to cite popular works in a microscopic generative model of scientific citation networks. The model allows us to understand how some papers are cited while others are forgotten in the context of a growing citation network. Moreover, it allows us to estimate an expected citation landscape of the future, predicting that 95% of papers relevant in 2050 have not yet been written. We identify some conditions under which individual publications manage to achieve immortality.

The impact of a disaster on a country or a community is often measured using one aggregate metric: the total cost of the physical damages. While relevant to estimate financial needs for the reconstruction, this single number hardly represent the impact on the poorest people and households, who suffer disproportionally from disaster but, because they own very little, experience little financial damages. This presentation will propose a different approach to measure the severity of disasters, based on microsimulations in which disaster impacts are represented at the household level. The presentation will use examples from multiple countries and disasters to illustrate the results and their policy implications. Open questions on which more research (and collaborations across disciplines) are needed will also be discussed.

The thermodynamic uncertainty relations (TURs) provide upper bounds on the precision of an arbitrary current in a system in terms of the entropy production (EP) of that system. All TURs derived so far have concerned a single physical system, varying in the assumptions they make about the dynamics of that system. However, many physical scenarios of interest involve multiple interacting systems, e.g. organelles within a biological cell. Here we show how to extend the previously derived TURs to those scenarios. A common feature of these extended versions of the TURs is that they bound the global EP, jointly generated by the set of interacting systems, in terms of a weighted sum of the precisions of the local currents generated within those systems – plus an information-theoretic correction term. We also exploit these extended TURs to obtain bounds that do not involve the global EP, but instead relate the local EPs of the individual systems and the statistical coupling among the currents generated within those systems. We derive such bounds for both scalar-valued and vector-valued currents within each system. We illustrate our results with numerical experiments.

The role of schools in the transmission of SARS-CoV-2 is still controversial. It is often hard to delineate transmissions in schools from transmissions involving school-aged children in other settings as well as to properly assess the impact of school-specific mitigation measures. Here, we tackle these challenges by analyzing extensive Austrian contact tracing data from 616 outbreaks with transmissions in schools involving 2,822 students and 676 teachers to calibrate an agent-based epidemiological model in terms of the distribution of cluster sizes and the dependence of transmission risks on age and symptomatic versus asymptomatic disease courses. Within this model we quantify the impact of different prevention measures including ventilation, class size reductions, wearing masks during the lessons, as well as screening strategies by means of antigen tests. While 40% of the clusters contained no more than two cases, we find that 3% of all clusters contained more than 20 infections. The younger the children, the more likely we found asymptomatic cases and teachers being the index case of the cluster. Different types of schools require different combinations of measures to achieve control of the virus spread (each case infects on average less than one other person). In primary schools, it is in general necessary to combine two of the above measures whereas, in secondary schools, where contact networks of students and teachers become increasingly large and dense, a combination of three measures is needed. A sensitivity analysis indicates that outbreak sizes might increase up to three-fold in secondary schools for virus variants with 50% increased transmissibility, whereas poorly executed or enforced mitigation measures might increase outbreak sizes by a factor of more than 30. Our results suggest that school-type-specific combinations of measures allow for a controlled opening of schools even under sustained community transmission of SARS-CoV-2, even though outbreaks involving multiple cases are to be expected on an infrequent but regular basis.

This talk will explore the ambiguous notion of “connectedness,” which, in the Panarchy framework, is one of the three dimensions through which complex systems move over time. This dimension is intimately linked with another dimension of complex systems – resilience. Thus, by measuring connectedness we can better understand and predict the resilience of complex systems. But how does one measure, or even conceptualize the connectedness of a complex system? Our approach is to use network theory, coupled with co-occurrence models from ecology, to elucidate and quantify the cryptic economic structures within cities. Taking that structure as a network we then develop a system-level measure of connectedness, which we find is positively correlated with both vulnerability and productivity of a city. This network approach can further be used to develop a typology of cities based on their responses to shocks. One goal of this work is to deliver novel insights and models to economic planners and policy makers, particularly in areas of resilience, innovation, and future well-being.

Research careers are typically envisioned as a single path in which a scientist starts as a member of a team working under the guidance of one or more experienced scientists and, if they are successful, ends with the individual leading their own research group and training future generations of scientists. Here we study the author contribution statements of published research papers in order to explore possible biases and disparities in career trajectories in science. We used Bayesian networks to train a prediction model based on a dataset of 70,694 publications from PLoS journals, which included 347,136 distinct authors and their associated contribution statements. This model was used to predict the contributions of 222,925 authors in 6,236,239 publications, and to apply a robust archetypal analysis to profile scientists across four career stages: junior, early-career, mid-career and late-career. All three of the archetypes we found – leader, specialized, and supporting – were encountered for early-career and mid-career researchers. Junior researchers displayed only two archetypes (specialized, and supporting), as did late-career researchers (leader and supporting). Scientists assigned to the leader and specialized archetypes tended to have longer careers than those assigned to the supporting archetype. We also observed consistent gender bias at all stages: the majority of male scientists belonged to the leader archetype, while the larger proportion of women belonged to the specialized archetype, especially for early-career and mid-career researchers.

Articles in high-impact journals are, on average, more frequently cited. But are they cited more often because those articles are somehow more “citable”? Or are they cited more often simply because they are published in a high-impact journal? Although some evidence suggests the latter the causal relationship is not clear. We here compare citations of preprints to citations of the published version to uncover the causal mechanism. We build on an earlier model of citation dynamics to infer the causal effect of journals on citations. We find that high-impact journals select articles that tend to attract more citations. At the same time, we find that high-impact journals augment the citation rate of published articles. Our results yield a deeper understanding of the role of journals in the research system. The use of journal metrics in research evaluation has been increasingly criticized in recent years and article-level citations are sometimes suggested as an alternative. Our results show that removing impact factors from evaluation does not negate the influence of journals. This insight has important implications for changing practices of research evaluation.

The aim of this project is to further our understanding of multimorbidity, the co-occurrence of two or more diseases in patients. Multimorbidity is associated with decreased quality of life, increases in hospitalizations, health care utilization, mortality, and costs of care. Therefore, the primary motivation behind this research project is the optimization of treatment paths by providing new insights into how multimorbid patient health states develop in men and women. The first question is a population-wide analysis of disease correlations and disease progression. Concerning this question, I will work on developing comorbidity networks of human diseases based on medical claims data considering age, sex, and time. The next question is the identification of sex-specific comorbidities. A structured the approach of identifying sex and gender-specific differences in comorbidities will be introduced.

Complexity is a phenomenon that various sciences are interested in. In the context of organizations, complexity refers to the degree of differentiation among the elements of an organization. Scholars have investigated, among others, the effects of organizational Complexity on innovation or organizational performance, in general. Increasingly, organizing takes place around information technology, thereby creating large quantities of digital trace data. Digital trace data provides a processual perspective on organizational complexity and thus allows scholars to investigate how organizational complexity develops as organizations grow, re-organize, or even cease to exist. This dissertation makes use of this emerging opportunity and investigates how organizational complexity changes over time using digital trace data from two organizations. In this regard, I pursue two complementary perspectives on organizational complexity: structural complexity (i.e. how an organization is organized) and process complexity (i.e. how an organization carries out work). With this doctoral dissertation, I extend computational methods to investigate organizationalcomplexity and provide further insights into how structural and process complexity change in the light of time.

In this talk, Miguel Fuentes will discuss some results concerning generalized non-linear dynamics equations and how these can give light to descriptions and interpretations of evolution processes, as well as the implications for diversification/innovation in biological systems. He will include some epistemological topics related to the evolution of scientific theories and complexity.

Sector-wise input-output tables are widely used to model the inter-industry dependencies of a national economy and how it responds to shocks, such as the economic recovery from disasters and pandemics. In conventional input-output analyses producers react to such shocks exclusively by adjusting the quantity of their produced goods only while prices remain fixed. Here, we propose an agent-based extension to these dynamic input-output recovery models in which producers can adjust their economic behaviour to external shocks by price and quantity adjustments simultaneously. The model is self-consistent in the sense that its production functions fulfill all economic constraints of the original Leontief input-output model at each time step of the dynamics. After the economy has absorbed the shock, the model allows for a continuum of different equilibrium configurations depending on whether producers are more likely to react by adjusting their produced quantities or the price of their products. To illustrate the model, we fit and evaluate it for the Austrian economy and how it absorbed the shock of the financial crisis in 2007/2008.

I will give a summary on and provide insights to scientific activities we designed and are still engaged into understand the dynamics of the ongoing COVID-19-pandemic. These activities include *) the application of computational models that we used to predict the global dissemination of the virus during the early phase of the pandemic, *) the discovery of universal sub-exponential growth regimes in the first epidemic wave in China and other countries and a theoretical model that explains these observations by accounting for systematic behavioral changes in the population, *) a nationwide high-resolution mobility monitor that we developed in Germany, that quantifies how much and in what way mobility was affected during lockdown periods, and *) a participatory experiment that we launched in March 2020 that involves >500,000 participants that donate resting heart rate, physical activity and sleep data by means of their smartwatches and based on which we designed and implemented a national fever thermometer to detect and predict the time course of confirmed cases of COVID-19 in Germany and that is now used as an important surveillance data stream for the federal COVID-19 situation analysis.

Will a large economy be stable? Building on Robert May’s original argument for large ecosystems, we conjecture that evolutionary and behavioural forces conspire to drive the economy towards marginal stability. We study networks of firms in which inputs for production are not easily substitutable, as in several real-world supply chains. We argue that such networks generically become dysfunctional when their size increases, when the heterogeneity between firms becomes too strong, or when substitutability of their production inputs is reduced. At marginal stability and for large heterogeneities, we find that the distribution of firm sizes develops a power-law tail, as observed empirically. Crises can be triggered by small idiosyncratic shocks, which lead to “avalanches” of defaults characterized by a power-law distribution of total output losses. This scenario would naturally explain the well-known “small shocks, large business cycles” puzzle, as anticipated long ago by Bak, Chen, Scheinkman, and Woodford.

The COVID-19 pandemic has exposed the world's population to sudden challenges that elicited strong emotional reactions. While investigations of responses to tragic one-off events exist, studies on the evolution collective emotions during a pandemic are missing. We analysed the digital traces of emotional expressions in tweets during the five weeks after the start of outbreaks in 18 countries and six different languages. We observed an early strong upsurge of anxiety-related terms in all countries, while sadness terms rose and anger terms decreased when casualties increased and social distancing measures were implemented in the following weeks. Positive emotions remained relatively stable. Our results show some of the most enduring changes in emotional expression observed in long periods of social media data, displaying a memory pattern supported by social interaction. This kind of time-sensitive analyses of large-scale samples of emotional expression have the potential to inform mental health support and risk communication.

This paper assesses the network robustness of the technological capability base of 269 European metropolitan areas against the potential elimination of some of their capabilities. By doing so it provides systematic evidence on how network robustness conditioned the economic resilience of these regions in the context of the 2008 economic crisis. The analysis concerns calls in the relevant literature for more in-depth analysis on the link between regional economic network structures and the resilience of regions to economic shocks. By adopting a network science approach that is novel to economic geographic inquiry, the objective is to stress-test the technological resilience of regions by utilizing information on the co-classification of CPC classes listed on European Patent Office patent documents. Findings from a regression analysis indicate that metropolitan regions with a more robust technological knowledge network structure exhibit higher levels of resilience with respect to changes in employment rates. This finding is robust to various random and targeted elimination strategies concerning the most frequently combined technological capabilities. Regions with high levels of employment in industry but with vulnerable technological capability base are particularly challenged by this aspect of regional economic resilience.

Due to its high lethality amongst the elderly, nursing homes are in the eye of the COVID-19 storm. Emerging test procedures, such as antigen or RT LAMP tests, might enable us to protect nursing home residents by means of preventive screening strategies. Here, we develop a novel agent-based epidemiological model for the spread of SARS-CoV-2 in nursing homes to identify optimal preventive testing strategies to curb this spread. The model is microscopically calibrated to data from actual nursing homes in Austria, including the networks of social contacts of their residents and information on past outbreaks. We find that the effectiveness of preventive screenings depends critically on the timespan between test and test result, the detection threshold of the viral load for the test to give a positive result, and the screening frequencies of residents and employees.

Automatically detecting semantic shifts (i.e., meaning changes) of single words has recently received strong research attention, e.g., to quantify the impact of real-world events on online communities. These computational approaches have introduced various measures, which are intended to capture the somewhat elusive and undifferentiated concept of semantic shift. On the other hand, there is a longstanding and well established distinction in linguistics between a word’s paradigmatic (i.e., terms that can replace a word) and syntagmatic associations (i.e., terms that typically occur next to a word). In this work, we join these two lines of research by introducing a method that captures a measure’s sensitivity for paradigmatic and/or syntagmatic (association) shifts. For this purpose, we perform synthetic distortions on textual corpora that in turn induce shifts in word embeddings trained on them. We find that the Local Neighborhood is sensitive to paradigmatic and the Global Semantic Displacement is sensitive to syntagmatic shift in word embeddings. By applying the newly validated paradigmatic and syntagmatic measures on three real-world datasets (Amazon, Reddit and Wikipedia), we find examples of words that undergo paradigmatic and syntagmatic shift both separately and at the same time. With this more nuanced understanding of semantic shift on word embeddings, we hope to analyze a similar concept of semantic shift on RDF graph embeddings in the future.

Colexification is a linguistic concept that has recently been used to assess the semantic similarity of words. Unfortunately, such assumption lacks validation at scale. In her talk, Anna considers one form of meaning — affective meaning — and shows that colexification networks follow such relationships in the creation of links between words. Moreover, Anna explores the possible applications of such properties, in particular the unsupervised expansion of pre-existing affective lexica and the interpolation of affective meaning of words. She then compares the performance of the network method with a state of the art machine learning corpus method.

Workers that become automated may transfer productivity gains to their co-workers or make it easier to automate their jobs too. In this paper, I empirically investigate how automatable jobs have diffused impacts to neighbouring jobs in North American cities between 2007 and 2016. Results indicate that jobs that share similarities with neighbouring high-risk jobs grew less, even when controlling for their own technical risk of automation. Conversely, jobs that share complementarities with neighbouring high-risk jobs grew faster, possibly indicating productivity gains from working with recently automated jobs. In addition to the analysis in this paper, I provide an adjusted index of job automation risk that accounts for local diffusion of impacts (negative and positive) in US cities.

Gender biases exist in a wide variety of human behavior. Examples are mobility, communication and leisure activities. Mobile phone datasets allow to investigate human behavior on very large study populations without introducing any observation bias. By studying mobile phone usage data in Austria during the COVID-19 crisis we are able to quantify gender differences in response to stress and crisis in a huge natural experiment. The dataset allows us to quantify changes in mobility, communication and sleep patterns. We find that existing gender biases in mobility and communication are increased during the COVID-19 pandemic. Males generally return to normal much quicker and the effects are not distributed evenly across age cohorts. For both genders, but even stronger in females, we observe a drastic increase in call duration. These findings relate strongly to existing literature, implying that females cope more actively with stress and males are more prone to risk taking.

The last decade saw the rise of deep learning. The abundance of available data and ever-increasing computational power pushed the boundaries of machine learning. Deep learning has been applied in many different areas and achieved state-of-the-art performance in many of them. However, privacy concerns are hindering the adoption of deep learning in fields like medicine or finance, even though it also could make a positive impact in these fields. This talk will present some methods for processing data in a privacy preserving way and show how it is possible to learn something about a given dataset without learning too much about the individuals in this dataset.

Bitcoin as a major cryptocurrency has come up as a shooting star of the 2017 and 2018 headlines. After exploding its price twenty times just in the twelve months of 2017, the tone has changed dramatically in 2018 after major price corrections and increasing concerns about its mining power consumption and overall sustainability. The dynamics and interaction between Bitcoin price and its mining costs have become of major interest. Here we show that these two quantities are tightly interconnected and they tend to a common long-term equilibrium. Mining costs adjust to the cryptocurrency price with the adjustment time of several months up to a year. Current developments suggest that we have arrived at a new era of Bitcoin mining where marginal (electricity) costs and mining efficiency play the prime role. Forecasting models building on this dynamics suggest that even though the hashrate will keep increasing (as well as Bitcoin price), it will be controlled by increasing mining efficiency and reward halving. In effect, the maximum profitable electricity price will keep decreasing forcing the use of cheap/renewable sources of energy. Environmental catastrophe induced by Bitcoin mining thus does not seem likely to threaten us.

Abstract: Our society relies on open source software in a fundamental way. Yet we often take this digital infrastructure for granted and ignore potential risks. In this talk I will discuss and attempt to quantify systemic risks present in open source ecosystems. I focus on social risks, highlighting that a few key individuals play a critical role in these decentralized systems. I will conclude by discussing potential solutions and improvements.

While deep networks are a highly successful model class they require substantial computation resources for their training, storage, and inference, which limits their effective use on resource-constrained devices. Many recent research activities explore different options for compressing and optimizing deep models for the Internet of Things (IoT). Our work so far has focused on two important aspects of deep neural network compression: class-dependent model compression and explainable compression. In this talk, we shortly discuss why these aspects are important for real-world applications and summarize our contributions.

Abstract: We live in a digital world today. Almost every aspect of our existence is recorded somewhere, on some computer. But, are these data on our behavior being used truly effectively? Ethically? Usefully? In this talk, we will cover a range of advanced technologies that are attempting to do exactly that. https://mirsadhadzikadic.academia.edu/cv

In my master thesis I analysed datasets that describe specific phases in the development of mouse embryos. We used methods from physics in combination with models of complex systems to gain new insights into the temporal structure of gene regulation during such developmental processes. The data sets consist of about 400 cells and the RNA expression profiles of a few thousand of their genes. The data samples the embryonic development of mice capturing the transition from one cell type to another one. The inference problem consists of two parts. The first part (i) is to infer a developmental order of cells in the sample. This part of the thesis is concerned with dimensional reduction and constructing a so-called pseudo-time description for the developmental process. The second part (ii) deals with inferring regulatory dynamics from the inferred pseudo temporal order of cells. We assumed that a developmental process can be described by a linearized differential equation with a non-linear constraint forcing the species abundances to remain positive. Finally, we compared both approaches and identified the genes that have the most impact on the process.

The availability of transaction-level payment and loan data for an entire nation gives us a unique opportunity to study the economy as a whole. We investigate the interaction of its agents, and the potential risks and opportunities arising from each economic agents decisions. The general objective of this project is to identify the supply chain of the Brazilian economy. This presentation is an opportunity to open the table for discussion and specially collect questions and ideas that can be answered by this unique kind of data.

Entropy is one of the most important concepts in thermodynamics and statistical physics. Boltzmann defined it as the logarithm of state multiplicity. The multiplicity is typically described by multinomial factors leading to the exponential growth of the sample space. In the chemical reaction we have free particles and molecules that are created from these particles. This leads to a super-exponential multiplicity. The typical approach to modelling the chemical reaction is through the grand canonical ensemble with several particle reservoirs such that the number of particles is conserved in average. However, it is possible to model the chemical reactions more correctly by properly counting the multiplicity of the emerging molecule states. This will be demonstrated on the basis of the example given by Jensen et al. [2018] called the Magnetic coin model.

This talk will present the vision of self-adjusting networks: communication networks whose physical topology adapts to the traffic pattern it serves, in a demand-aware manner. Such networks are enabled by emerging reconfigurable optical technologies. It will be shown that the benefit of self-adjusting networks depends on the amount of “structure” there is in the demand, and an information-theoretical approach to measure the complexity of traffic traces will be presented to derive entropy-based metrics accordingly. Optimal offline and online algorithms to design self-adjusting networks whose performance matches the derived metrics asymptotically will be discussed. Bio: Stefan Schmid is a Professor at the Faculty of Computer Science, at University of Vienna, Austria. He obtained his diploma (MSc) in Computer Science at ETH Zurich in Switzerland (minor: micro/macro economics, internship: CERN) and did his PhD in the Distributed Computing Group led by Prof. Roger Wattenhofer, also at ETH Zurich. As a postdoc, he worked with Prof. Christian Scheideler at the Chair for Efficient Algorithms at the Technical University of Munich and at the Chair for Theory of Distributed Systems at the University of Paderborn, in Germany. From 2009 to 2015, Stefan Schmid was a senior research scientist at the Telekom Innovation Laboratories (T-Labs) and at TU Berlin in Germany (Internet Network Architectures group headed by Prof. Anja Feldmann). In 2013/14, he was an INP Visiting Professor at CNRS (LAAS), Toulouse, France, and in 2014, a Visiting Professor at Université catholique de Louvain (UCL), Louvain-la-Neuve, Belgium. From 2015 to 2017, Stefan Schmid was a (tenured) Associate Professor in the Distributed, Embedded and Intelligent Systems group at Aalborg University, Denmark, and continued working part-time at TU Berlin, Germany. Since 2015, he serves as the Editor of the Distributed Computing Column of the Bulletin of the European Association of Theoretical Computer Science (BEATCS), since 2016 as Associate Editor of IEEE Transactions on Network and Service Management (TNSM), and since 2019 as Editor of IEEE/ACM Transactions on Networking (ToN). Stefan Schmid received the IEEE Communications Society ITC Early Career Award 2016. His research interests revolve around the fundamental and algorithmic problems of networked and distributed systems.

Management of systemic risk in financial markets is traditionally associated with setting (higher) capital requirements for market participants. There are indications that while equity ratios have been increased massively since the financial crisis, systemic risk levels might not have lowered, but even increased (see ECB data1; SRISK time series2). It has been shown that systemic risk is to a large extent related to the underlying network topology of financial exposures. A natural question arising is how much systemic risk can be eliminated by optimally rearranging these networks and without increasing capital requirements. Overlapping portfolios with minimized systemic risk which provide the same market functionality as empirical ones have been studied by Pichler et al. (2018). Here, a similar method for direct exposure networks is proposed, and applied to cross-sectional interbank loan networks, consisting of 10 quarterly observations of the Austrian interbank market. We show that the suggested framework rearranges the network topology, such that systemic risk is reduced by a factor of approximately 3.5 (70%), and leaves the relevant economic features of the optimized network and its agents unchanged. The presented optimization procedure is not intended to actually re-configure interbank markets, but to demonstrate the huge potential for systemic risk management through rearranging exposure networks, in contrast to increasing capital requirements that were shown to have only marginal effects on systemic risk (Poledna et al., 2017). We compute that on average bank equity needs to be increased by a around 230% to achieve the same reduction in DebtRank as our optimization procedure. Ways to actually incentivize a self-organized formation toward optimal network configurations were introduced in Thurner and Poledna (2013) and Poledna and Thurner (2016). For regulatory policies concerning financial market stability the knowledge of minimal systemic risk for a given economic environment can serve as a benchmark for monitoring actual systemic risk in markets. Systemic risk in financial networks has been extensively investigated for different types of contractual networks. Examples are interbank loan networks, derivative exposure networks, equity cross holding networks or exposure networks arising from asset crossholdings of agents. More recently the paradigm of multi-layer networks has been used to model systemic risk for multiple financial networks jointly. [1] aggregated different direct exposure networks and showed that looking at single layers in isolation can underestimate systemic risk drastically. [2] take a different approach using extensions of classical centrality measures, like eigenvector centrality or page rank to multiplex networks. However, these algorithms cannot provide a monetary quantification of banks systemic riskiness in case of defaults. DebtRank, which aggregates the losses occurring in asset valuation cascades does not suffer this problem, but is only defined on a single layer [3,4]. Thus, we suggest to extend the valuation shock transmission mechanism of DebtRank into the multilayer framework when considering networks layers, which cannot be aggregated by summation. We establish such an extension of DebtRank by incorporating the funding liquidity contagion layer into the algorithm. Thus, this allows us to quantify the costs for the financial system in case of bank defaults in a more complete way. It also allows us to simulate different combinations of funding liquidity shocks and valuation shocks jointly. Since, valuation and funding liquidity cascades are interacting on most bi-layer network specifications we expect higher systemic risk impacts of bank defaults, than if the two contagion mechanisms are modelled separately. In a related study [5] modelled these contagion channels as supraadjacency matrix in order to analyse network stability properties for small simultaneous macro shocks. Considering multiple networks simultaneously can change the structure of higher order connection of agents. In financial transaction networks this means that second and higher order risks agents are facing when creating a link, can look dramatically different in comparison to a single layer perspective. The multilayer extension of DebtRank can take such higher order connections arising from different layers into account. References 1. Poledna, Sebastian, et al. “The multi-layer network nature of systemic risk and its implications for the costs of financial crises.” Journal of Financial Stability 20 (2015): pp 70-81. 2. Bardoscia, Marco, Ginestra Bianconi, and Gerardo Ferrara. “Multiplex network analysis of the UK OTC derivatives market.” (2018). 3. Battiston, Stefano, et al. “DebtRank: Too Central to Fail? Financial Networks, the FED and Systemic Risk.” Scientific reports 2 (2012): 541. 4. Bardoscia, Marco, et al. “DebtRank: A Microscopic Foundation for Shock Propagation.” PloS one 10.6 (2015): e0130406. 5. Wiersema, Garbrand, et al. “Inherent Instability: Scenario-Free Analysis of Financial Systems with Interacting Contagion Channels.” (2019).

Is it the same a reliable transmission than a copy? This question, apparently superficial, turns out to be strikingly deep, and shakes the foundations of information theory: No measure explicitly accounting for the amount of bits copied in a given information exchange is part of the standard body of information theory. However, a general inspection of the modes of information transfer tells us that information can be transmitted in two qualitatively different ways: by copying or by transformation. Copying occurs when messages are transmitted without modification, e.g., when an offspring receives an unaltered copy of a gene from its parent. Transformation occurs when messages are modified systematically during transmission, e.g., when mutational biases occur during genetic replication. Standard information-theoretic measures do not distinguish these two modes of information transfer, although they may reflect different mechanisms and have different functional consequences. Starting from a few simple axioms, I will show a decomposition of mutual information into the information transmitted by copying versus the information transmitted by transformation. In addition, I will show how copy information can be interpreted as the minimal work needed by a physical copying process, which is relevant for understanding the physics of replication. The presented results apply to any system in which the fidelity of copying, rather than simple predictability, is of critical relevance. This includes genetic replication, animal communication, unsupervised machine learning or the evolution of artificial codes.

Text analysis methods are used in many applications. Hand-crafted databases have been used by both linguistics scholars and data scientists to develop new analysis methods. Although they have been proved to work for extracting meaning, analyzing emotions, performing disambiguations and many other applications, manual and supervised methods lack scalability, while unsupervised methods often are not properly validated or capture patterns without linguistic interpretation. I will introduce and discuss a novel method to analyze texts that uses only translation databases. The method is based on the linguistic phenomenon of colexification, identifying relations between words through their translations.

In September 2019, the Seshat Global History Databank began a long-term collaboration with the Complexity Science Hub, forming a new Research Group on Social Complexity and Collapse. This unique databank serves as a massive repository of structured data about the evolution of past societies. It captures information on sociopolitical organization, religion, warfare, technology, and agriculture for a global sample of societies from the Neolithic to the Industrial Revolution. Jenny Reddish, Lead Editor for the project, will give a brief introduction to how Seshat works, how we’re trying to bridge the gap between the sciences and the humanities, and what our research goals are for the future.

This talk presents an application in collaborative recycling efforts for non-profit agencies. The work is motivated by a project with a network of hunger relief agencies (e.g., food pantries, soup kitchens and shelters) focusing on collaborative approaches to address their cardboard recycling challenges collectively. The problem is modeled as a periodic location routing problem with operational choice. A metaheuristic is compared to the exact solution of the problem. The relative difficulty introduced with each decision is examined. Computational results on synthetic instances as well as a case study based on data from the network will be presented. In this novel setting, we evaluate collaboration in terms of participation levels and cost impact. These insights can be generalized to other networks of organizations that may consider pooling resources.

The talk will discuss knowledge diffusion and knowledge coordination in modern economies. Knowledge, or better know-how, is an important determinant of economic prosperity and growth. With a growing global body of know-how, but more or less constant capacity of individuals to master it, economies face the problem that know-how becomes more and more distributed across people and teams of people. Tapping and coordinating this know-how is therefore crucial to the success of complex economies. The talk will present a number of examples of how large data sets, such as social security, cell phone and payment card data, can be used to explore the diffusion and coordination of know-how, highlighting the role of migration, foreign investments, business travel and cities in a globalizing economy. Bio: Frank Neffke is the Research Director of the Growth Lab at the Harvard Kennedy School. His research focuses on economic transformation and growth. He has written on a variety of topics, such as structural transformation and new growth paths in regional economies, economic complexity, division of labor and teams, the consequences of job displacement and the future of work. Before joining the Growth Lab, Frank worked as an assistant professor at the Erasmus School of Economics in Rotterdam, The Netherlands.

We study opinion dynamics on a network of friendships and enmities, where agents' opinions and the relationships among them coevolve, although on separate timescales. We model the agents opinions and their relationships as binary spins and coupling constants, respectively. The dynamics of both spins and couplings are implemented via the Metropolis algorithm, assuming they are coupled to heat baths at different temperatures. For this coevolutionary dynamics, at equilibrium, we find a continuous transition from an ordered phase, where agents are partitioned into two antagonistic groups to a mixed phase, where the bipartition disappears and opinions are randomly distributed amongst agents. If only the agents' relationships change over time while the opinions remain unchanged, then, in addition to the disappearance of the network bipartition, we observe a spin-glass-like transition from a unique network configuration to multiple possible configurations.

Polarization is threatening the stability of democratic societies. Until now, research has focused on the extremeness aspect of polarization, and failed to explain the correlation between different polarized issues, i.e., \emph{hyperpolarization}. We develop an opinion dynamics model based on psychological balance theory that, for the first time, is able to generate hyperpolarization, and to explain the link between affective and opinion polarization. We validate the predictions of this model on empirical data from the 2016 National Election Survey.

Great public speakers are made, not born. Practicing a presentation in front of colleagues is common practice and results in a set of subjective judgements what could be improved. In this paper we describe the design and implementation of a mobile app which estimates the quality of speaker’s delivery in real time in a fair, repeatable and privacy-preserving way. Quantle estimates the speaker’s pace in terms of the number of syllables, words and clauses, computes pitch and duration of pauses. The basic parameters are then used to estimate the talk complexity based on readability scores from the literature to help the speaker adjust his delivery to the target audience. In contrast to speech-to-text-based methods used to implement a digital presentation coach, Quantle does processing locally in real time and works in the flight mode. This design has three implications: (1) Quantle does not interfere with the surrounding hardware, (2) it is power-aware, since 95.2 % of the energy used by the app on iPhone 6 is spent to operate the built-in microphone and the screen, and (3) audio data and processing results are not shared with a third party therewith preserving speaker’s privacy. We evaluate Quantle on artificial, online and live data. We artificially modify an audio sample by changing the volume, speed, tempo, pitch and noise level to test robustness of Quantle and its performance limits. We then test Quantle on 1017 TED talks held in English and compare computed features to those extracted from the available transcript processed by online text evaluation services. Quantle estimates of syllable and word counts are 85.4 % and 82.8 % accurate, and pitch is over 90 % accurate. We use the outcome of this study to extract typical ranges for each vocal characteristic. We then use Quantle on live data at a social event, and as a tool for speakers to track their delivery when rehearsing a talk. Our results confirm that Quantle is robust to different noise levels, varying distances from the sound source, phone orientation, and achieves comparable performance to speech-to-text methods.

Networked systems are ubiquitously present in the real world – food webs, inter-bank markets or communication networks, for instance. Prediction of collapses in networked systems is an extremely important, still almost impossible task. The main obstacle is typically the enormous size of the network resulting in extremely large amount of information necessary to describe its structure. Moreover, in many systems the network structure remains hidden. Based on the corollary of the famous Perron-Frobenius theorem called eigenvector quantization, we show that for a broad class of networked systems it is possible to detect the last stage before the crash without knowing the network structure and consequently to predict the collapse of the whole system.

Recent projections indicate that the number of connected Internet-of-Things (IoT) devices, sensors, and actuators will pass 46 billion in 2021 and people will interact with these devices on a daily basis, e.g., to monitor and control our health, home, connected vehicles, security systems, social interactions, and almost every other aspect of our daily activities. As one of the most natural ways of communication, speech has recently found a rapidly increasing interest as primary mode of interaction between humans and IoT devices. At the same time, recent research has shown that there are clear links between the neurological and mental wellness of an individual and patterns in the individual's speech. Timely detection of such impairments can help improve the user-machine interactions (e.g., adapt the computing systems to the mental or cognitive conditions of the user) and prevent problems (such as operation of safety-critical equipment by users lacking appropriate cognitive fitness) before they cause damage to humans and machines. In this talk, I will discuss several challenges in continuous and non-intrusive speech assessment and present our ongoing research efforts in several specific case studies. Bio: Christian Poellabauer received his Dipl. Ing. degree from the Vienna University of Technology, Austria in 1998 and the Ph.D. degree from the Georgia Institute of Technology, Atlanta, GA in 2004, both in Computer Science. He is an Associate Professor in the Department of Computer Science and Engineering at the University of Notre Dame and was awarded a 2019 Fulbright Scholar grant to pursue his academic interests at TU Graz. His research interests are in the areas of wireless sensor networks, mobile computing, ad-hoc and vehicular networks, pervasive computing, and mobile healthcare systems. He has published over 150 scientific contributions in these areas and he has co-authored a textbook on Wireless Sensor Networks. His research has received funding through the National Science Foundation, the National Institutes of Health, Department of Education, Department of Defense, IBM, Intel, Toyota, Ford Research, Motorola Labs, National Geographic, the National Football League, GE Health, and several other foundations and businesses. He received the Outstanding Dissertation Award from Georgia Tech and an NSF CAREER Award in 2006. He is a senior member of ACM and IEEE.

Multimorbidity, the co-occurrence of two or more chronic diseases such as diabetes, obesity or cardiovascular diseases in one patient, is a frequent phenomenon. To make care more efficient, it is of relevance to understand how different diseases condition each other over the life course of a patient. However, most of our current knowledge on such patient careers is either confined to narrow time spans or specific (sets of) diseases. Here, we present a population-wide analysis of long-term patient trajectories by clustering them according to their disease history observed over 17 years. When patients acquire new diseases, their cluster assignment might change. A health trajectory can then be described by a temporal sequence of disease clusters. From these cluster transitions we construct an age-dependent multiplex network of disease clusters. Random walks on this multiplex network provide a more precise model for the time evolution of multimorbid health states when compared to models that cluster patients based on single diseases. We find that for elderly patients the cluster network consists of two different regions, one for clusters with low and one with high in-hospital mortality. Our results can be used to identify the crucial events that potentially determine the future disease trajectory of a patient.

We analyzed a dataset of general social media text-based expression: 22 Million status updates donated by 153727 Facebook users. By applying a lexicon-based approach, we quantified emotions in each post in terms of valence, the degree of pleasure associated with an emotional experience, and arousal, the level of activity induced by the emotional experience. We further replicated our analysis with the unsupervised VADER method and supervised valence and arousal methods specific for Facebook. We analyzed how the emotions of a post predict the emotional content of the next post by the same user as a function of time between posts ?t. It can be observed that emotional expression relaxes quickly but not instantly, as changes in short timescales are directed towards the baseline but do not fully converge. We captured this dynamics through a dynamic equation model following an agent-based modelling framework for emotions in online interaction. This way we modelled emotion eigendynamics as an example of an Ornstein–Uhlenbeck process. We fitted the solution of this model against both datasets through nonlinear regression and came up with three main results: i) the presence of emotional expression indicates emotion regulation, evidencing a correction of emotional intensity as soon as a message is written, ii) valence and arousal decay exponentially towards their baselines, and iii) the baseline of valence is above its midpoint but the baseline of arousal is slightly below its midpoint (on a 1-9 scale). The above results are in line with well-established dynamics of emotions from previous research in psychology, but our analysis has several advantages. We observe a much longer period (for Facebook on average 534.36 days) over a larger population than the largest previous works with self-reports. Additionally, we capture general emotion dynamics rather than changes due to emotion labelling: This enables us to quantify individual dynamics that inform future computational models and analyses of emotional expression in social media.

Human society relates through different sorts of interactions that are not independent of each other. The question arises whether these interdependencies can be used to predict the dynamical behaviour of such relationships. Here we look for empirical evidence that the dynamic behaviour of a given layer of a social system is related to the behaviour of other layers. Succeeding in this endeavour would make possible to improve predictions. But it would also improve the understanding of humanity as a set of interacting elements and the social dynamics between these elements. The Massive Multiplayer Online Game (MMOG) Pardus is the source of data. In this game, players live in a virtual, futuristic universe where they interact with each other in a multitude of ways to achieve their self-posed goals. Such system represents a complete human society. If the interaction between two people is represented as a link between two nodes, the whole set of human interaction can be described as a multiplex network. In such a system, layers represent different sorts of interactions. One basic phenomena in a network is the formation and disappearance of links over time. A key question is whether it is possible to predict link appearances in one layer of the multiplex system using knowledge about the other layers. To answer this, the well-known problem of link prediction is generalized by extending it to the multiplex scenario represented by our social system.

Lying is traditionally defined as stating something that is believed to be false with the intention that the other person believes it to be true. However, there are cases where we believe something to be false only to an extent and we can nevertheless lie about it. We call this partial lying. We provide an epistemological analysis of these situations and investigate how the severity of partial lying may be determined, and how partial lies can be classified. To make the investigation more realistic, we also study how much epistemic damage an agent suffers depending on her level of trust that she invests in the liar and the degree of belief in the falsehood of the liar's statements. The results are demonstrated through computer simulations of an arguably rational Bayesian agent who is trying to determine how biased a coin is while observing the coin tosses and listening to an (un)trustworthy person's misleading predictions about the outcomes. Our results provide an interesting insight at the intersection of epistemology and ethics, namely that in the longer term partial lies lead to more epistemic damage than outright lies. However, if the liar only lies sometimes, then categorical lies are more epistemically damaging as the liar is considered to be more trustworthy.

Socialdata often comes from a heterogeneous population composed of non-randomly sampled subgroups, each with different characteristics and behaviors. A population-level trend in the aggregate data may disappear or reverse itself when the same data is disaggregated into its underlying subgroups. This effect, known as Simpson’s paradox, confounds analyses of social data, including inference of trends and causal effects. I illustrate the problem with several examples showing how the paradox can distort conclusions of analysis, and describe recent algorithmic efforts to address this problem. Our algorithm systematically disaggregates data to identify subgroups whose behavior deviates significantly from the rest of the population. The method allows us to leverage Simpson’s paradox to uncover interesting patterns in real-world social data, such as Q&A site Stack Exchange and online learning platforms Khan Academy and Duolingo. Bio: Kristina Lerman is a Principal Scientist at the University of Southern California Information Sciences Institute and holds a joint appointment as a Research Associate Professor in the USC Computer Science Department. Trained as a physicist, she now applies network analysis and machine learning to problems in computational social science, including crowdsourcing, social network and social media analysis. Her recent work on modeling and understanding cognitive biases in social networks has been covered by the Washington Post, Wall Street Journal, and MIT Tech Review.

The amount of medical data generated in healthcare has exponentially increased in recent years. To gain a quantitative understanding of the complex processes involved in healthcare, we need new scientific methods to adequately investigate the heterogeneous and dynamic data they produce. Data-driven analysis, when applied in a clinical context, has the aim to turn complex medical data records into knowledge on how to prevent or treat diseases more effectively. This approach requires the combination of data-related, methodological expertise with in-depth know-how of the involved clinical or medical domain. More precisely, it requires interdisciplinary collaboration, since data analysis competencies needs to be consistent with medical domain expertise. However, successful interdisciplinary research remains a great challenge. In order to transform medical data into knowledge, we need to gain new insights into factors that support interdisciplinary interaction and the emergence of cognitive processes through which disciplinary concepts and methods are integrated. Therefore, this master thesis has two objectives. First, to carry out a data-driven analysis by means of an interdisciplinary research team and, second, to combine our practical experience with theoretical knowledge about cognitive processes to gain knowledge about interdisciplinary research. In terms of methodological data analysis approaches, we applied multiple logistic regression to analyse the dosage-dependent occurrence of osteoporosis in statin patients. This study reveals that there is a highly non-trivial dependence of statin dosage with the odds of osteoporosis. To the best of our knowledge, this is the first study which shows that it is important to consider both potency and dosages when investigating the relationship of osteoporosis and statin therapy. Our results show that the diagnosis of osteoporosis is underrepresented in low-dose and overrepresented in high-dose statin treatment. The second high-level objective was to show how interdisciplinarity promotes data-driven analysis and to gain insights into factors that support the interaction process between researchers from different disciplines and the emergence of a cognitive process among them. We made use of contemporary cognitive science theories and Bourdieu’s theory of practice to get an alternative perspective on interdisciplinarity. Based on these theoretical insights combined with our own research experience, we identified body—mind--environment actors that we consider important for a successful research process: personal/habitus-related qualities, environmental/field-related properties, a shared study logic and time. In summary, this work demonstrates how data-driven analysis is promoted by interdisciplinary research.

Modern macroeconomic theories were unable to foresee the last Great Recession and could neither predict its prolonged duration nor the recovery rate. They are based on supply--demand equilibria that do not exist during recessionary shocks. Here we focus on resilience as a nonequilibrium property of networked production systems and develop a linear response theory for input--output economics. By calibrating the framework to data from 56 industrial sectors in 43 countries between 2000 and 2014, we find that the susceptibility of individual industrial sectors to economic shocks varies greatly across countries, sectors, and time. We show that susceptibility-based predictions that take sector- and country-specific recovery into account, outperform---by far---standard econometric growth models. Our results are analytically rigorous, empirically testable, and flexible enough to address policy-relevant scenarios. We illustrate the latter by estimating the impact of recently imposed tariffs on US imports (steel and aluminum) on specific sectors across European countries.

The emergence of ubiquitous sensing and control of the physical world through networks of embedded computing devices has currently gone beyond conventional areas of ambient and environmental monitoring and reached the cost-sensitive, generally risk averse and safety critical industrial domain. Data-driven models have been deployed for process fault detection and diagnosis, improving yields of production lines and optimisation of key performance indicators such as energy efficiency and maintenance and productivity related metrics, with outreaching social and environmental impact. The talk aims to discuss the current context and challenges related to the design, implementation and evaluation of advanced data processing and learning algorithms in industry applications. Special focus is given to in situ efficient inference on distributed embedded devices by means of novel deep neural network architectures. Bio: Grigore Stamatescu is Associate Professor at the University Politehnica of Bucharest, Romania and a Visiting Research Scientist at the Complexity Science Hub and the Technical University of Graz. His current research interests span the areas of networked embedded sensing, the internet of things and distributed information processing in industry, the built environment and smart city applications, with more than 100 publications in international journals and conference proceedings. He was a 2015-2016 Fulbright Visiting Scholar and he is a current recipient of the Joint Excellence in Science and Humanities grant of the Austrian Academy of Sciences. Grigore Stamatescu is member of the IEEE Robotics and Automation Society, ACM and the IFAC TC 3.3. Telematics: Control via Communication Networks.

Technological change and industrial organization are connected on many levels: Where do new technologies emerge?, how do they diffuse in the industry and among consumers?, what impact does this have on the characteristics of the cross-sectional distribution of firms? Technologies do not evolve isolated: They rely on other technologies, on infrastructure, and on specialized knowledge. This causes network externalities and links the future success of technologies to the number of previous users of not only these technologies, but also of the pre-existing infrastructure and related technologies. Observing the emergence and diffusion of technologies is crucial for our understanding of both the role and potential of technological change and our capacity to incentivize the emergence of green technologies, encourage prudent practices in finance and insurance, and address problems such as climate change. Some results from agent-based modeling and replicator dynamics are presented and substantiated by empirical results from patenting in green technologies, as well as industrial dynamics in the insurance business and other sectors.

The aim of this talk is to revise the SWIR model which has recently been used as a general framework for various cascading dynamics, such as k-core percolation, breakdown on interdependent networks and cooperative epidemic spreading. While this model works well for static networks, the adaptive nature of many real-world processes requires to consider the network evolution as well. It will be discussed how to incorporate an adaptation into the original SWIR to uncover new phenomena that do not appear over static networks.

The nature of statistics, statistical mechanics and consequently the thermodynamics of stochastic systems is largely determined by how the number of states W(N) depends on the size N of the system. Here we propose a scaling expansion of the phasespace volume W(N) of a stochastic system. The corresponding expansion coefficients (exponents) define the universality class the system belongs to. Systems within the same universality class share the same statistics and thermodynamics. For sub-exponentially growing systems such expansions have been shown to exist. By using the scaling expansion this classification can be extended to all stochastic systems, including correlated, constraint and super-exponential systems. The extensive entropy of these systems can be easily expressed in terms of these scaling exponents. Systems with super-exponential phasespace growth contain important systems, such as magnetic coins that combine combinatorial and structural statistics. We discuss other applications in the statistics of networks, aging, and cascading random walks.

Adaptive co-evolving dynamic networks play a key role in ecological, epidemiological, social or financial systems. In adaptive network models the node states and the network topology are dynamically coupled, i.e. they co-evolve. These models have in common that they can collapse. We would like understand these collapse transitions and their precursors in adaptive network models. Structural information about the network is of crucial importance for this task. These models differ strongly with respect to the level of aggregation at which this information is required. On one side there are systems whose collapse depends solely on a very particular network configuration and on the other side there are systems whose collapse depends on aggregated structural variables, such as the overall density of certain network motifs. In this talk I contrast two representative models that differ in this respect: The Jain-Krishna model and the adaptive SIS model. In the first model the collapse occurs when cycles are broken. We show that this collapse can be predicted accurately purely from information of the nodes, even though it relies crucially on a particular network topology. In the second model we show that a prediction may be systematically off, even in the presence of aggregated knowledge of the network topology.

Data analysis and data mining in particular as emerging fields of data science have gained wide popularity in the last years both in academia and in industry. As an example, recent research initiatives such as Industry 4.0, SmartFactory or Internet-of-Things (IoT) try to motivate researchers and engineers to improve the production and products in various application fields by utilizing technologies for data analysis such as clustering, filtering and visual data analysis for example. Beside many issues that have to be solved in this intent, the major problem that still remains here is how to deal with the growing amount of data/information, i.e., a so-called information overload problem. Visual data analysis has been proven to be one of the effective ways to tackle this issue. Visualizations have a distinctive advantage when dealing with the information overload problem: because they are grounded in basic visual cognition, many people understand them and can naturally perform visual operations such as clustering, filtering and comparing quantities. However, creating appropriate visualizations requires specific expertise of the domain and underlying data. Yet, an ordinary user lacks expert knowledge and can rarely generate sophisticated visualizations. Thus, the first quest of this work is to provide strategies to automatically recommend appropriate visualizations for non-experts by following visual encoding rules and perceptual guidelines. Yet, considering just visual encoding rules leads to a large set of possibilities, valid in terms of representing the data visually, but without considering which type serves the users' needs best. To tackle this issue, we propose a novel recommender system that (i) recommends visualizations based on a set of visual cognition rules and (ii) filters a subset considering the user's preferences. This research work investigates different strategies to recommend visualizations considering different aspects of the user preferences/needs/interest. Bio: Belgin Mutlu is a senior researcher at the Know Center GmbH and area manager at the Pro2Future GmbH in Graz. She received her Master’s and Bachelor’s degree in Telematics, and her PhD degree in Informatics from Graz University of Technology. Her research interests include visualizations, visual data analysis, recommender systems and semantic web. Belgin has worked as researcher and developer in the field of adaptive visualizations, linked data and visual recommender systems in several EU Projects—CODE, EEXCESS, and AFEL. She has co-authored more than 15 peer-reviewed publications.

Affective disorders such as depression, anxiety disorder or PTSD are characterized by a symptom called ‘negative repetitive thinking’: Patients’ thoughts tend to circle around a small set of topics. Conversely, in a state of psychosis, patients often experience ‘flight of thought’, i.e. their thoughts leap from topic to topic, often following only very distant associations. How can we develop metrics to detect these symptoms in spoken or written communication? I discuss and compare several approaches, such as word/text embeddings models, and measures derived from large-scale word association networks. I will also give an outlook on other potential application areas, such as creativity research.

New Keynesian Dynamic Stochastic General Equilibrium (NK-DSGE) models with numerous real and nominal frictions have become the standard theory-based model for economic forecasting. To explain the origins of these models in general equilibrium theory, this talk will give an overview of the theory and assumptions behind NK-DSGE models. Secondly, it will explain how these models are taken to the data using Bayesian estimation techniques, and present results of a forecast evaluation exercise with a two-country NK-DSGE model for Austria and the Euro Area. Lastly, the lecture reflects on the continued use and apparent success of NK-DSGE models in economic forecasting, despite widespread critique of their underlying theory and the way Bayesian estimation is conducted in a DSGE context.

We are used to working with independent ensembles or systems exhibiting detailed balance. If the typical system we are interested in are not of this kind then what happens with our mathematical tools to describe such systems? And can we say anything about how irreversible a process works? In this lecture we will briefly discuss these two questions. For the first question we briefly look at the maximum configuration entropy of driven processes. For the second question we will look more closely at thermodynamic cycles performed in finite time in a gas kinetic model to find a relation between the length of cycle periods and the amount of irreversible work obey a relation similar to a Heisenberg uncertainty relation.

In this talk, we will present our last results on the modelling of rumour and disease spreading in single and multilayer networks. We will introduce a general epidemic model that encompasses the rumour and disease dynamics into a single framework. The susceptible-infected-susceptible (SIS) and susceptible-infected-recovered (SIR) models will be discussed in multilayer networks. Moreover, we will introduce a model of epidemic spreading with awareness, where the disease and information are propagated in different layers with different time scales. We will show that the time scale determines whether the information awareness is beneficial or not to the disease spreading. Finally, we will show how machine learning can be used to understand the structure and dynamics of complex networks.

In today’s smart world, the consumer products and electronics, sensors, industrial components, and other everyday objects are getting increasingly integrated and connected through the Internet backbone, and thereby enabling emerging application scenarios in the world of Internet-of-Things (IoT). The massive amount of data generated by the diverse interrelated heterogeneous devices (typically embedded systems) has to be first computed, transferred in real time via internet and take actions without requiring human or computers contact/intervention. This poses high demands on adaptability/intelligence, energy-efficiency and reliability for these systems that require advanced features to enable correct functionality and availability of the services to the user despite sudden interruptions in the system operation, device failures, harsh environments, and escalating security threats. This talk will start with a quick overview of important dependability issues, prominent state-of-the-art techniques, and various hardware/software modeling and optimization techniques developed by me together with my research team. A key focus will be on bridging the gap between hardware and software to achieve accurate reliability models at the higher system layers while accounting for the underlying hardware features. This provides a foundation to develop and employ diverse robustness optimizations at different system layers of (heterogeneous) computing devices deployed at the Edge, under given performance and power constraints. I will also give a quick outlook of my recent works in approximate computing that allow trading output quality/accuracy with efficiency in terms of area, latency, power, and energy consumption, thereby enabling highly energy-efficient and low-latency Edge nodes. Towards the end, I will present a summary of my other professional activities and research proposals, followed by an outlook of my future vision and research agenda.

Innovation can often be formulated as a recombinatorial-process where objects from a set are combined to form new objects that are in turn added to the set of objects that can again be recombined to form yet new objects. In many instances this is a social process where the recombination is performed by a group of collaborators. This is especially true when it comes to the creation of new knowledge or methods as this mostly happens by people exchanging and combining their current knowledge and ideas to create new knowledge. For this to happen communication between the collaborators is necessary. However in social psychology one can often observe a phenomenon called Ringelmann-Effect, which states that as such groups grow in size the increasing coordination overhead reduces the rate of successful communications. This ultimately leads to less innovative exchanges and a decline in novelty production. In this talk I want to give a short overview about a simple combinatorial innovation model that mimics the recombination of knowledge elements by a set of agents and an analysis of data from Open-Source-Software projects. The aim is to investigate the influence of the Ringelmann-Effect in the model as well as in the data-set.

In this talk I will first explain the goals and sub domains in visualization research, followed by a short overview of some of the most compelling publications presented at this year's EuroVis 2018 conference.

If average global warming is to be contained below 2°C, greenhouse gas emissions from the energy sector must be eliminated by mid-century. Wind and solar photovoltaics provide a large untapped resource and have experienced rapid cost reductions. They are thus poised to form the backbone of a clean energy system. However, questions remain about how to integrate them into a stable, affordable and sustainable energy system at scales ranging from cities to continents, in particular due to their weather-dependency, the resulting requirements to spread them spatially, and the additional technologies like electricity storage needed for their integration. This presentation will provide an overview of ongoing work to model these technologies and their integration into the energy system, and the key open challenges to successfully manage the renewable energy transition.

Web-based phenomena such as fake news, hate speech or echo chambers have been gaining increasing attention in the public discourse. However, the extent and actual impact of these phenomena are still unknown. In order to address these issues in more depth, it is important to understand how opinions emerge, change and get exchanged. In our work, we therefore intend to develop novel, data-driven approaches to examine social influence mechanisms and opinion forming at a large scale. We aim, in particular, to introduce a complex user model, which characterizes each person with respect to three levels: 1) the individual level capturing the characteristics of a user; 2) the network level capturing the social relationships of a user; and 3) the group level describing sets of similar users with certain dominant opinions and high-level ideologies. For the empirical analysis we use a rich and unique dataset of an online news forum that contains the complete posting history of all users, their detailed behaviors and interactions as well as all news articles and the respective meta data for almost two decades. Short Bio: Julia Neidhardt is a researcher at the division of E-Commerce at TU Wien. She holds a Master's degree in mathematics from the University of Vienna and a PhD in Computer Science from TU Wien. Her research focuses on modeling and predicting complex human behavior, user preferences and social relations as well as their dynamics in digital-enabled environments. Current projects are concerned with social influence in online communities, the emergence and diffusion of topics, opinions and sentiments, picture-based travel recommender systems, and social media-based event prediction. Since 2013, she has been regularly visiting the Science of Networks in Communities (SONIC) research group at Northwestern University, USA.

The availability of large scale healthcare data promises to revolutionise medical care towards a precision medicine where treatments are tailored individually to each patient. In the last decade, the analysis of medical records, often recorded for billing purposes, was centred around binary relations between diseases, such as the comorbidity between them. The concept of comorbidity quantifies the probability that a patient with disease A also has disease B, but does not capture the long term history of a patient. For example, the subsequent development of diseases A,B and C could indicate the presence of a particular genetic defect, increasing the likelihood for disease D. The aim of our work is to bridge this gap by developing a method to forecast future diagnoses of a patient which takes into account the entire history, i.e., the specific sequence, of his or her diagnoses. To this end, we analyse a data set containing for each inpatient hospital stay of patients in Austria over two years the main and side diagnoses and the admission and release dates. The data thus contains a trajectory on the set of diseases for each of the approximately 8M people insured via an Austrian social security provider. We represent the health history of each patient by a binary vector, and divide the set of patients into categories by using a hierarchical clustering algorithm. When patients acquire new diseases, they can change their cluster. We analyse the pattern of transitions between the different clusters, in particular, we see that there exists a stream of patients into clusters acting as sinks, characterised by highly multimorbid patients.

Human symbol systems such as art and fashion styles emerge from complex social processes that govern the continuous re-organization of modern societies. They provide a signaling scheme that allows members of an elite to distinguish themselves from the rest of society. Efforts to understand the dynamics of art and fashion cycles have so far been based on costly signaling theory, where elite members signal their superior status by introducing new symbols (e.g. fashion-styles), which are subsequently adopted by low-status groups. In response to this adoption, the elite members need to introduce yet new symbols to signal their status. We propose an alternative explanation based on counter-dominance-signaling. There, members of the elite want others to imitate their symbols; changes only occur when outsider groups successfully challenge the elite by introducing signals that contrast those endorsed by members of the elite. To clarify the mechanism that actually drives fashion cycles in musical styles, we use a dynamic network approach on data containing almost 8 million musical albums released between 1956 and 2015. There a network systematically quantifies artistic similarities of competing musical styles. By studying the dynamics of the network we can formulate statistical hypothesis tests for whether new symbols are introduced (i) by current elite members as predicted by costly signaling theory or (ii) as a consequence of challenges by peripheral groups through countersignals. We find clear evidence that counter-dominance-signaling drives changes in musical styles. This provides a quantitative, completely data-driven answer to a century old debate about the nature of the underlying social dynamics of art and fashion cycles.

Affective Intelligence Theory (AIT) postulates that political behavior is strongly influenced by affective states. Enthusiasm and aversion lead people to rely on simple cognitive heuristics and reinforce social identity, whereas anxiety causes people to adopt more complex cognitive strategies and weakens group boundaries. Until now, AIT has mostly been tested on survey data. Instead, we tested a dimensional reformulation of AIT (DAIT) against data from a German-speaking online news community, 20min.ch. In this community, users discuss recent news and give up and down votes to each others' comments, often producing heated discussions. We find that discussions characterized by low affective potency and extreme valence ("enthusiasm" and "aversion") produce more polarized votes and signal lower cognitive complexity than discussions with high potency ("anxiety") and/or less extreme valence. This effect is even stronger when there is a high salience of social identity in discussions. Thus, we can confirm DAIT, and open the door for further analysis of DAIT in other languages and settings.

1) I will discuss the consistency of Maximum entropy principle for generalized entropies (e.g. Tsallis or Rényi entropy). Recently, several authors discussed the statistical consistency of general entropies with different results. During the talk, we show that the class of entropies obeying the consistency axioms is relatively large and enables us to model systems with inherent correlations. 2) Superstatistics is a tool which enables us to model processes on different scales assuming the existence of local equilibria. In the original definition are only two scales present. We show how to generalized the superstatistics to arbitrary number of characteristics scales, leading into transition of superstatistics for different (time)-scales. We also demonstrate the transition on the example of high-frequency data from financial markets.

Measuring the efficiency of a road network is currently an open question and many approaches have been suggested. In this paper we tackle this issue by proposing a network generation model, studying its properties and laying out a definition for the performance of a network. We then confront the real road network against the highest performance version of this model and define efficiency as the ratio between the performance of the real road network and our idealised model. This is obtained through the definition of local and weighted versions of centrality measures. These measures deal with distance-decay effects and nodes having different masses. In doing so, we remark that each possible network created with our model has a typical fractal dimension, and this permits to map a fractal dimension (a way to occupy space) with a performance.

The adaptive SIS model is one of the most elementary models in which the dynamics on the network is coupled to the dynamics of the network. We investigate the critical behavior of the network topology by looking at the densities of motives, the clustering, the compactness, the degree distribution and assortativity, the effective branching ratio and spectral distribution. The proximity to the persistence threshold can be sensed in some of these quantities long before the transition via the appearance of local maxima due to a crossover phenomenon. This is also reconcilable with the pair approximation. A good signal-to-noise ratio makes them novel candidates for precursors of the transition. We explain the mechanism that gives rise to this behavior and some caveats. We also discuss the issues that arise when fitting power laws to these quantities.

“More is different,” wrote P.W. Anderson [1] in order to characterize the behaviour of complex systems, consisting of many interacting elements, which cannot be explained by a simple extrapolation of the laws describing the behaviour of a few elements. In order to understand complex dynamics and emergent phenomena we need to understand interactions among system components. Interacting subsystems can mutually exchange information and influence each other; or one system can causally influence another one by a directed information flow. The mathematical formulation of causality in measurable terms of predictability was given by the father of cybernetics N. Wiener [2] and formulated for time series by C.W.J. Granger [3]. The Granger causality is based on the evaluation of predictability in bivariate autoregressive models. This concept has been generalized for nonlinear systems using methods rooted in information theory [4, 5]. Complexity observed in such systems as the human brain or the Earth climate stems not only from the fact that they consist of many subsystems. Their variability covers large ranges of spatial and temporal scales and the nonlinear character of these systems leads to interactions of dynamics across scales. For instance, dynamical processes on large time scales influence variability on shorter time scales. In order to detect cross-scale causal interactions we have recently introduced a methodology [6] which starts with a wavelet decomposition of a multi-scale signal into quasi-oscillatory modes of a limited bandwidth, described using their instantaneous phases and amplitudes. Then their statistical associations are tested using the information-theoretic formulation of the Granger causality. The analysis of long-term air temperature records uncovers causal influence and information transfer from large-scale modes of climate variability with characteristic time scales from years to almost a decade to regional temperature variability on short time scales. The phenomenon of cross-scale interactions has non-negligible influence on the air temperature variability in the European mid-latitudes [7]. In the tropical Pacific, the interactions of the annual cycle with slower modes of atmospheric and oceanic variability are of special interest in order to better understand the El Niño Southern Oscillation. The cross-frequency interactions and information transfer can be observed in brain dynamics, where the cross-frequency coupling enriches the cooperative behaviour of neuronal networks and apparently plays an important functional role in neuronal computation, communication, and learning. [1] P. W. Anderson, Science 177, (1972) 393 [2] N. Wiener, in: E. F. Beckenbach (Editor), Modern Mathematics for Engineers (McGraw-Hill, New York, 1956) [3] C.W.J. Granger, Econometrica 37 (1969) 424 [4] K. Hlavá?ková-Schindler et al., Phys. Rep. 441 (2007) 1 [5] M. Paluš, M. Vejmelka, Phys. Rev. E 75 (2007) 056211 [6] M. Paluš, Phys. Rev. Lett. 112 (2014) 078702 [7] N. Jajcay, J. Hlinka, S. Kravtsov, A. A. Tsonis, M. Paluš, Geophys. Res. Lett. 43(2) (2016) 902–909

Scholars have long debated the existence and properties of a “common structure” that underlies narratives. Using modern, computer-based language analysis methods, we measured several structural and psychological categories of language across multiple corpora of novels, short stories, films, and written projective stories. Across all forms of stories, analyses revealed a small but consistent underlying structure to narratives that is driven by 3 primary processes: Staging, Plot Progression, and Cognitive Tension. This research provides empirical insights into the structure of narrative, providing hints as to what it may tell us about the evolution of story-making.

Polarization poses an existential threat to democratic political systems. Until recently, polarization was analyzed purely in terms of political positions. However, it becomes more and more apparent that polarization must also be seen as an affective phenomenon. In this presentation, we show that affect 1) correlates with polarization on the macro-level of Swiss political history, quantified with a novel measure of relational polarization, 2) increases the repetitiveness of online discussions, quantified on the basis of a neuro-probabilisitic language model, and 3), facilitates the alignment of issue positions. We also present an agent based model that explains the coupling between affect and opinion alignment, and produces testable psychological hypotheses. Our outcomes suggest that affect is not only an epiphenomenon, but a causal driver of political polarization.

Multi-agents systems (MAS) have been introduced in the field of artificial intelligence (AI) to denote a set of entities which act and interact in intelligent (i.e., often: rational) ways. Beyond AI, the application of results concerning MAS involves various other scientific disciplines, such as: biology; sociology; finance; and economics. In the literature, a large number of models have been proposed to describe the dynamics of MAS. Most of these are based on simulations, limiting their validity to the particular scenarios considered and to the specific choice of simulation parameters: In order to derive more robust results, it is of interest to identify viable analytic approaches. Our framework is inspired by mathematical kinetic theories, with the aim to obtain analytic results on the dynamics of MAS from the description of the effects of single interactions among agents. KTMAS can incorporate different types of interactions, and it can be used to describe the dynamics of different features of MAS. We present our research in the scope of the new disciplines of econophysics, which can be used to model wealth evolution, and sociophysics, which can be adopted to characterise opinion evolution.

In this talk I will present a fresh overview of my past projects on human behavior with the emphasis on agent-based modeling. Among the topics I will focus on the project on inferring human mobility using communication records. Our results show that human mobility is well predicted by a simple model based on the frequency of mobile phone calls between two locations and their geographical distance. We argue that the strength of the model comes from directly incorporating the social dimension of mobility. Furthermore, as only aggregated call data is required, the model helps to avoid potential privacy problems.

The availability of large scale healthcare data, promises to revolutionise medical care, with treatments tailored individually to each patient. In the last decade, the analysis of medical records, often recorded for billing purposes, was centered around binary relations, such as the comorbidity, between two diseases. Comorbidity quantifies the probability that a patient with disease A also has disease B. While this measure is often sufficient for practical purposes, in other cases it may be beneficial to consider the long term history of a patient, that is the interplay of more than two diseases, and their temporal order. For example, the simultaneous presence of diseases A,B and C could indicate the presence of a particular genetic defect, increasing the likelihood for disease D. To our knowledge, the time sensitive analysis of interactions of more than two diseases has not yet been pursued systematically in the literature. The aim of our work is to bridge this gap by developing a method to predict future diagnoses of a patient which takes into account the entire history of the patient's diagnoses. To this end, we analyse a data set containing for each inpatient hospital stay in Austria the ID of the patient, the main and side diagnoses and the admission and release dates for the years 2006 and 2007, and the same information for Lower Austria for the years between 2008 and 2011. The data thus contains a "disease trajectory" for each of the around 8M people insured via an Austrian social security provider. This data can be combined with data for health care provider (e.g. GP, dermatologist, ...) visits and data on medical prescriptions, which is available to us for the same patient cohort. In this talk I will explain our current ideas on how to tackle the problem of diagnosis prediction beyond pairwise statistics.

In this talk we will discuss the catalytic network model that Jain and Krishna introduced in 1998 as a toy-model of evolution. This model exhibits cycles of growth, organisation and collapse. After pointing out that it is much more generic than expected, we will exhibit some of the interesting questions that pop out of its rich structure: Can one anticipate the length of the organised phase or predict the time and strength of a collapse? To this end we will make a short excursion to the statistics of binary threshold tests and then tackle some questions regarding the prediction of catastrophic events in this model.

A recent path to understand the dynamics of evolutionary systems is “combinatorial evolution". The essence of this approach is, that an evolving system can be formulated as a set of objects, which can be combined to pairs to form new objects that are in turn added to the set and can be again combined to form yet new objects and so forth. In this talk I want to give a short overview about a simple combinatorial evolution model how knowledge in the form of novel ideas is created through the recombination of preexisting ideas. I especially want to focus on how the performance of the system depends on its size and draw parallels to how the efficiency of groups in producing new knowledge might depend on the size of the group.

Air pollution is traditionally monitored by networks of static measurement stations operated by official authorities. These stations are highly reliable and accurately measure a wide range of air pollutants with expensive analytical instruments. Large size, high price, and laborious maintenance of these complex measurement systems severely limit the number of installations resulting in a low spatial resolution of the available pollution information. In the last few years, low-cost solid-state gas sensors became available on the market. We integrated these sensors in ten lightweight air quality monitoring stations and deployed them on top of trams in Zurich to increase spatial coverage of the urban area. In this talk I will focus on three main challenges we faced in this deployment: 1) pre-deployment sensor testing and compensation for cross-sensitivities, 2) automatic calibration of low-cost sensors to improve data quality, and 3) data analysis and predictions to construct high resolution air pollution maps. I will give an overview of the theory behind the developed methods and evaluate their performance on a data set comprising 400 million measurements collected in Zurich over three years. Generated pollution maps received a lot of attention in the local media and are now part of the public service offered by the City of Zurich.

This paper describes a novel approach for estimating the indirect economic losses due to natural disasters with an application to flood events in Austria. The approach combines a probabilistic physical damage catastrophe model with a macroeconomic agent-based model (ABM). The ABM methodology advances state-of-the-art approaches by exploiting large data sets (big data) with close to a million agents (households, non-financial and financial firms and a general government) calibrated with data from national accounts, input-output tables, government statistics, census data and business information. The probabilistic flood model is equally innovative by introducing a copula methodology that provides an assessment of flood losses by taking account of spatial dependencies in the flood hazard. The ABM includes an input-output model with 64 industries where all goods and services are produced endogenously, and the probabilistic copula approach provides a nation-scale estimate of direct flood losses over the full risk spectrum based on basin-scale loss distributions and exposure with Corine land-cover mapping. The direct loss estimates are used to build a damage scenario generator that provides the input for the ABM, which, in turn, assesses the indirect economic losses due to the event. The analysis shows that there can be severe indirect economic losses in Austria due to large-scale natural disasters, or systemic events, and shows the impact chains leading to the systemic losses. One of the main findings is that the distinct types of disaster events exhibit qualitatively different economic behavior: while more moderate scenarios induce positive indirect economic effects in the medium term, the severe, or systemic, events result predominantly in a negative economic response throughout the simulation period. Most importantly, the analysis disaggragates the gains and losses occurring to different sectors. This detailed information can be useful for assessing risk management options at various scales.

Collective emotions are emotional states temporarily shared by large amounts of individuals. From riots to sport events, from viral content to online quarrels, collective emotions are subject to appear in various situations and have long-lasting consequences. Understanding collective emotions has been challenging due to their fast evolution, large scale, and complex dynamics, limiting their tractability in natural scenarios and their controllability in experiments. Recent developments in computational social science, in particular agent-based modeling and sentiment analysis of digital traces, allow us to quantify and model emotions at unprecedented scales and resolutions, offering new opportunities to study collective emotions. I will present an overview of modeling and analysis of collective emotions at scale, with an emphasis on how these can be observed through social media and how they can be reproduced in computational models. I will illustrate the application of digital traces of collective emotions to understand the collective responses to the Paris terrorist attacks of November, 2015. Using a large-scale dataset of Twitter public messages, we measured affect as expressed through text. We find the traces of the simultaneous negative reaction to the attacks, which are later replaced by a synchronization process that increases the salience of shared values and prosocial behavior. Our analysis supports the hypothesis that synchronization leads to social resilience after a collective trauma, illustrating the social function of collective emotions.

We replicate Saha and Gounder (Econ Model 31: 70-79, 2013), who show that the relationship between corruption and economic development exhibits an inverted U-shaped pattern. Pooled OLS estimates, however, should not be interpreted as conclusive evidence in support of a causal relationship. Using a much longer panel of 156 countries during 1900-2010, we show that a U shape shows up even after controlling for country-level heterogeneity, suggesting that the estimated relationship does not result from a version of the classic “Kuznets fallacy”. The curvilinear relationship is also robust to controlling for potential simultaneity using an instrumental variable approach. Still, the parabola implied by a fixed-effects model is considerably less “peaked” than the one implied by the corresponding OLS model.

With the help of transfer entropy, we analyze information flows between communities of complex financial networks. We show that the transfer entropy provides a coherent description of interactions between communities, including non-linear interactions. We analyze transfer entropies between communities of five largest financial markets, represented as networks of interacting stocks. Additionally, we discuss information transfer of rare events, which can be analyzed by Rényi transfer entropy.

All living organisms, including humans, have internal biological or circadian clock that helps them anticipate and adapt to the regular rhythm of the day. The timings of human activities are marked by circadian clocks which in turn are entrained to different environmental signals. In an urban environment, the presence of artificial lighting and various social cues tend to disrupt the natural entrainment with the sunlight. However, it is not completely understood to what extent this is the case. Here we exploit the large-scale data analysis techniques to study the mobile phone calling activity of people in large cities to infer the dynamics of urban daily rhythms. From the calling patterns of about 1,000,000 users spread over different cities but lying inside the same time-zone, we show that the onset and termination of the calling activity synchronizes with the east-west progression of the sun. We also find that the onset and termination of the calling activity of users follows yearly dynamics, varying across seasons, and that its timings are entrained to solar midnight. Furthermore, we show that the average mid-sleep time of people living in urban areas depends on the age and gender of each cohort, most likely as a result of biological and social factors.

Systemic risk is the notion that a shock may not only render directly affected components but (possibly large) parts of the network non-functioning due to cascading effects. In a financial network this systemic risk stems from the exposures of lending institutes to one another. Network measures such as DebtRank were used to estimate the possible impact on the interbank network resulting from a market shock. In this work, I extend the analysis of systemic risk from financial institutes to the whole economy. and find that the DebtRank distribution of companies is qualitatively similar to the distribution of financial institutes. Companies may have exposures to a disadvantageous set of banks, thus enabling the propagation of an impact through large parts of the network. Even though larger companies tend to have a higher DebtRank, there are companies sharing a very similar DebtRank with total assets that span multiple orders of magnitude. The results suggest that from a systemic risk perspective companies and banks have some similarities. It may therefore be beneficial to extend regulations that try to alleviate systemic collapses to the whole economy.

We develop an agent-based model (ABM) for the Austrian economy using data from national accounts, input-output tables, government statistics, census data and business surveys. The model incorporates all economic activities (producing and distributive transactions) as classified by the European system of accounts (ESA) and all economic entities, i.e. all juridical and natural persons, are represented by agents (at a scale of 1:10). We show that this model is able to compete with vector autoregressive (VAR) and autoregressive–moving-average (ARMA) models in out-of-sample prediction. Potential applications of this ABM include economic forecasting, as well as the prediction of responses of the economy to endogenous shocks, e.g. from the financial system, or exogenous shocks like natural disasters, transformative technological innovations or unintended consequences of political interventions such as subsidies and tax policies.

Systemic risk arises as a multi-layer network phenomenon, where layers represent direct financial exposures of various types, including interbank liabilities, derivative- or foreign exchange exposures. Another network layer of systemic risk emerges through common asset holdings of financial institutions. Strongly overlapping portfolios lead to similar exposures caused by price movements of financial assets. We use a simple method to quantify systemic risk of overlapping portfolio networks from endogenous asset sales within the network. We then present a general optimization procedure where we minimize the systemic risk in a given financial market by optimally re-ordering overlapping portfolio networks, under the constraint that the expected returns and risks of the individual portfolios is unchanged. We explicitly demonstrate the method on the overlapping portfolio network of sovereign exposure between major European banks by using data from the European Banking Authority stress test 2016. We show that multiple systemic-risk-efficient allocations do exist, which are actually accessible by the optimization and that do not alter the returns and risks of the institutions’ portfolios. In the case of sovereign exposure, systemic risk can be reduced by more than 56%, without any detrimental effects. We confirm these results by a simple simulation of fire-sales in the government bond market.

How can one numerically simulate large stochastic networks? Perhaps one would go through each node in the network and update its state with a probability dictated by the denition of the model. This may not always be very ecient and can also lead to biases. The Gillespie algorithm provides a great alternative. I will present its merits and its implementation. The talk will close with the illustration of the Gillespie algorithm for an adaptive stochastic network dynamics, namely the SIS model with rewiring.

The Galilean composition law of velocities within Newtonian mechanics is additive. But, in order to unify mechanics with Maxwell electromagne+sm, Einstein adopted the Lorentz space-time transformation as the primary mathematical-physical goal to be satisfied. It resulted the well known, nonadditive, relativistic composition law of velocities. This surely is a small price to pay in order to unify mechanics and electromagnetism, and explain very many experimental facts. Analogously, there is a plethora of analytical, experimental, observational and computational evidences (see Bibliography in h]p://tsallis.cat.cbpf.br/biblio.htm) which reveals various kinds of violations of Boltzmann-Gibbs statistical mechanics, including thermodynamics. The adoption of nonadditive entropic functionals which generalize the traditional, additive, Boltzmann-Gibbs entropy enables to satisfy classical thermodynamics: as before, a small price to pay!

A wide range of physical, social and biological phenomena can be expressed in terms of spreading processes on networked systems. Notable examples include the spread of infectious diseases through direct contacts, the spatial propagation of epidemics driven by mobility networks, the spread of cyber worms along computer connections, or the diffusion of ideas mediated by social interactions. All these phenomena arise from a complex interplay between the spreading process and the network’s underlying topology and dynamics. Understanding how the time-evolving properties of the network impact the spread of the disease is a crucial step to setting up control and prevention strategies. The increasing availability of highly-resolved interaction data has made it possible to target a wide variety of settings and diseases, but at the same time new methodological challenges have arisen. In particular, a fundamental property of such phenomena is the presence of an epidemic threshold, i.e., a critical transmission probability above which large-scale propagation occurs, as opposed to quick extinction of the epidemic-like process. Computing this threshold is of utmost importance for epidemic containment and control of information diffusion. I will present a new analytical framework for the computation of the epidemic threshold for an arbitrary time-varying network. By reinterpreting the tensor formalism of multi-layer networks, this framework allows the analytical calculation of the epidemic threshold, without making any assumption on contact structure and evolution, and can be applied to a wide class of diseases. Along these theoretical developments, challenges related to the analysis and elaboration of the increased amount of data have emerged. I will present the case of diseases affecting farmed cattle. Such diseases compromise both human and animal health and welfare, and represent a major cause of loss in economic revenue. As a result, studying the networks of animal movements is a key step in devising new prevention and containment strategies. Past works have already analyzed cattle networks in several European countries. A comprehensive study, showing the impact of country-specific driving factors on network evolution and topology, is however still missing. I will present a collaborative platform for analyzing and comparing networks from several European countries. Using a bring code to the data approach, our platform overcomes the strict regulations preventing data sharing, and allows an effective comparative analysis. I will describe the framework, and present the result of this analysis, highlighting both properties that are characteristic of livestock markets, and country-specific features.

A particular class of linear master equation dynamics with unidirectional internal evolution rates and a long-jump resetting to the zero-state will be discussed. This model class is able to describe several distributions observed in complex systems, derived from the particular state dependence of the transition rates. We select out the exponential (geometrical) and the Waring distribution, along with their large system - continuous model counterparts, the Boltzmann-Gibbs and the Pareto distribution, as examples, respectively. A list of suitable properties for defining an entropic distance measure will also be presented.

In sustained growth with random dynamics, stationary distributions can exist without detailed balance. This suggests thermodynamical behaviour in fast growing complex systems. In order to model such phenomena, it is possible to use both a discrete and a continuous master equation. The derivation of elementary rates from known stationary distributions is a generalisation of the fluctuation–dissipation theorem. Entropic distance evolution is given for such systems. Depending on the transition and loss rates, the distributions for growing networks, particle production, scientific citations and income distribution can be reconstructed.

In different fields entropy has been conceptualized in different ways resulting in a functional expression equivalent to Shannon-entropy. Therefore entropy often is thought of as a universal concept. This however is true only as long as the underlying processes in question are essenially Bernoulli processes leading to a degeneracy of most commonly used entropy concepts and the ubiquitous emmergence of the particular functional form of Shannon-entropy. If the underlying process class in question becomes history dependent, also the expressions for the entropy functionals change and it becomes important which concepts we are using in order to define a notion of entropy; whether we think of an extensive property of matter, information rates, or maximum configuration principles, for instance, then starts to make a difference. For history dependent processes entropy is no longer uniquely defined and different entropy concepts have to be distinguished, informing us about distinct properties of the process. We discuss the general situation with simple history dependent example processes.

The society has a multi-layered structure, where the layers represent the different contexts resulting in a community structure with strong overlaps. To model this structure we begin with a single-layer weighted social network (WSN) model showing the Granovetterian correlations between link strength and topology. We find that when merging such WSN models, a sufficient amount of inter-layer correlation is needed to maintain these correlations, but they destroy the enhancement in the community overlap due to multiple layers. To resolve this, we devise a geographic multi-layer WSN model, where the indirect inter-layer correlations due to the geographic constraints of individuals enhance the overlaps between the communities and, at the same time, the Granovetterian structure is preserved. The network of social interactions can be considered as a multiplex from another point of view too: each layer corresponds to one communication channel and the aggregate of all them constitutes the entire social network. However, usually one has information only about one of the channels, which should be considered as a sample of the whole. We show by simulations and analytical methods that this sampling may lead to bias. For example, while it is expected that the degree distribution of the whole social network has a maximum at a value larger than one, we get with reasonable assumptions about the sampling process a monotonously decreasing distribution as observed in empirical studies of single channel data. We analyse the far-reaching consequences of our findings.

Economic and financial crises are extremely costly, yet our understanding of the resilience of economies is only in its infancy. Resilience is typically understood as a system’s ability to absorb, withstand, and also recover from adverse events. The characterization of economic resilience is particularly challenging due to strong interdependences between different industries and the way they absorb and recover from production shocks. In this talk we develop for the first time a quantitative, predictive, and data-driven formalism for the resilience of economies based on linear response theory (LRT). Key to this formalism is Leontief’s input-output model that depicts inter-industry relationships within an economy. We show how the impacts of production shocks in specific industries can be analytically derived from small-scale fluctuations observed in the input-output model using LRT. For each industry sector in 43 different countries, we can compute its time-dependent response to different kinds of shocks in any other sector. That is, a central result of this framework is the derivation of resilience curves for each sector with respect to shocks in any other sector. The so-derived resilience curves are shown to be predictive for, both, the average growth of and fluctuations in the output in a given country. Furthermore, we show that the impact of sector-specific demand shocks on the entire economy can also be understood and to some extent predicted from these curves. The predictions are shown to be particularly accurate when the framework is applied to the financial crisis in 2008. Our work establishes a firm and novel link between the out-of-equilibrium behavior of severely disrupted economies and their steady-state fluctuations.

A major problem for evolutionary theory is understanding the so called open-ended nature of evolutionary change, from its definition to its origins and consequences. Open-ended evolution (OEE) refers to the unbounded increase in complexity that seems to characterise evolution on multiple scales. In this talk we will present a fundamental characterisation of OEE. Essentially, it is assumed intrinsic unpredictability and the need for an always increasing amount of information to explain the successive evolutionary steps —the emergence of innovation. Interestingly, such unpredictability defines the boundary conditions for a mathematical problem which ends with a prediction: The statistical counterpart of the OEE ‘postulates’, based on standard Shannon Information theory, have the structure of a variational problem which is shown to lead to Zipf’s law as the expected consequence of an evolutionary processes displaying OEE. Interestingly, many complex systems candidates of displaying OEE, from language to proteins, share this common scaling behaviour. Other information-theoretic phenomena arising from open-ednedness, such as the paradox of information loss, will be also discussed. We will finish discussing the connection of this general framework with existing models for the understanding of the emergence of innovation.

The widespread use of internet allows to conduct experiments in the frame of social science with the engagement of a large number of participants. In this presentation I will show the results of a web-experiment intended to uncover voters' perception of the Italian political space in the early 2013, right before last Italian political elections. We find that political opinions follow a sort of Weber-Fechner law, known to occur in the perception of our five senses, i.e., our perceptions depend on the logarithm of stimuli intensity. Such finding can be exploited in the future to devise more realistic agent-based models of opinion dynamics.

A brief introduction to the concept of extortion strategies in iterated prisoners dilemma games and recent insights into the evolution of extortion and cooperation. Lit: (i) WH Press & FJ Dyson PNAS 109 (2012) and (ii) C Hilbe, MA Nowak & K Sigmund PNAS 110 (2013)

With the advent of social media it has become possible to study human social relations in a quantitative way. However, in most cases only data on positive relations (like friendship) are available while social balance theory states that in a social network positive and negative relations strongly depend on each other. In the massive multiplayer online game PARDUS players can mark each other not only as friends but also as enemies, leading to a two-layer multiplex network structure. We discuss the dynamics of friendship and enmity relations between thousands of players in PARDUS. We identify and quantify the driving processes of the associated two-layer social network formation. Well known sociological hypotheses like ‘The enemy of my enemy is my friend’ turn out to be important building blocks of understanding the dynamics of the coupled formation of friendly and hostile interactions within a society.

Optical imaging enables rapid in vivo imaging with micrometer scale resolution. In this seminar, I will present some recent technological advances and optical imaging applications in the eye and brain. In particular, we will focus on functional extensions of a technique called optical coherence tomography (OCT). OCT enables imaging tissue microstructure as well as polarization properties and perfusion in real time. Several applications for quantitative imaging of pathological changes in humans and small animals will be presented.

Content of my project work and the seminar presentation is to ascertain the probability of the presence of particles which move with Brownian motion in a complex, time-varying medium. Such transport processes are denoted as anomalous diffusion and can't be distinguished by the Fick's laws of diffusion. For such problems computer-aided algorithms of Brownian motion often lead to success, implemented as a numerical simulations of discrete random walker processes on networks. We are especially interested in anomalous diffusion on dynamic, complex networks, which alter over time. Propagation on medium can consist of diffusion or of iterative centrality measure such as the Katz prestige, eigenvector centrality or PageRank which all base on random walker processes with distinct behaviors. We try to illustrate the equivalence of (1) the simulation of random walkers, (2) the numerical calculation of the iterative equation and (3) the closed, analytical form of the diffusion processes. As an interdisciplinary background we use networks of clinical history of patients. The nodes of the network stand for the different phenotypes of diseases and the connecting link represents the likelihood that diagnosis appear together. These networks alter in age which is represented in a time-varying, dynamical network. Another main concern is to clarify how far trajectories and probability of presence of random walkers on network correlate with prevalences, evidence and precursor of diseases.

Different from traditional quantitative linguistics, we study languages in the following ways: distribution of the appearing frequencies in the text and emerging speed of new words. Usually these properties are studied for a book or a specific language, while seldom looks into the other aspects. Thanks to the Corpus of Historical American English(COHA), we get the full text corpus for about more than 110,000 texts ranging from 1810 to 2009. The texts are analyzed using the above approaches in terms of different sentence length. We observe patterns that stay invariant through history. This may indicate the invariance of the habits of language use. But more important, the invariant pattern indicates that the sentence length of 7 is sitting at a special point from the view of new word introduction rate and information density. This result is somehow linked to the results from the famous paper “The Magical Number Seven, Plus or Minus Two”.

The idea of “path” is fundamental to disentangle the complexity of the systems that are the outcome of an evolutionary process. The path described by an evolving system is often constraining the future space of alternatives, a property usually called history or path-dependence. In this talk I will show how history dependence is deeply linked with the scaling patterns observed in many complex systems, thanks to the recently developed theory of Sample Space Reducing (SSR) processes. From complex networks to critical phenomena, scaling laws emerge in somewhat regular way, and the comprehension of the mechanisms behind scaling patterns has become on of the hot topics of modern statistical physics. SSR process are a totally new route to scaling based on the unique assumption that the phase space reduces as long as the process unfolds. Applications of the theory of SSR processes to diffusive phenomena in complex networks, for example, reveal a source of unexpected results. I will finally revise some work-in-progress extensions of the theory which include i) Sample Space Reducing cascades: Multiplicative processes over a shrinking space, with applications from avalanche study to the energy spectrum of cosmic rays. ii) Sample Space Expanding (SSE) processes, a class of processes which can be understood to be the “mirror” of SSR processes. It is worth to mention that classic results of Record Statistics can be mapped into the simplest case of SSE processes.

Most disorders are caused by a combination of multiple genetic and environmental factors. If two diseases are caused by the same mechanism, they often co-occur in patients. Here we disentangle how much genetic or environmental risk factors contribute to the pathogenesis of 358 individual diseases, respectively. We pool data on genetic, pathway-based, and toxicogenomic disease-causing mechanisms with co-occurrences obtained from almost two million patients. From this data we construct a multilayer network where nodes represent disorders that are connected by links that either represent phenotypic comorbidity or the joint involvement of certain mechanisms. We quantify the similarity of phenotypic and mechanism-based links for each disorder. Most diseases are dominated by genetic risk factors, while environmental influences prevail for disorders such as depressions, cancers, or dermatitis. The relevance of environmental risk factors for a given disease is inversely related to its broad-sense heritability and also inversely related to the rate at which new drugs for the disease are approved. This might be indicative of a lack of successful drug development for diseases with high environmental risks. Our approach allows to rule out certain types of disease-causing mechanisms when their implied comorbidities are not observed and might therefore be used to identify promising leverage points for the development of future therapies of multifactorial diseases.

There are several different measures for the importance of the vertices in a network, often referred to as „Centrality“. I will compare different centrality measures, in particular the Debt-Rank, introduced by Battiston et al. to analyze the systemic risk in a financial network that can lead to a collapse as seen in the financial crises of 2007 - 2008. In our research we try to reconstruct the liabilities between Austrian companies and banks from aggregated data to estimate what companies could lead to a collapse of the economy in the case of a default.

The financial crises of 2007-2008 has led to increased awareness of the importance of systemic risk, which started pertinent contributions to quantifying the impact of inter-connectedness within financial networks and their interconnection to the real economy. Present research on financial networks and bank-firm networks attempts to estimate systemic risk that is generated by credits or other types of financial contracts. In this work we incorporate ownership-ties as an additional type of interconnection between entities. We generate an Austrian financial and real economy multi-layer network with two layers consisting of liabilities and shares held between and within banks and the real economy. We then quantify the contributions to systemic risk from each layer of this multi-layer network. To generate the network structure we use shareholder data from the Austrian commercial register and aggregated exposure data from banks to the real economy in Austria from the years 2006 to 2015.

Homophily, the tendency of individuals to associate with others who share similar traits, has been identified as a major driving force in the formation and evolution of social ties. In many cases it is however not clear if an observed homophily is the result of a socialization process, where individuals change their traits according to the dominance of that trait in their local social networks, or if it results from a selection process, in which individuals re-shape their social networks so that their traits match those in the new environment. Here we demonstrate the existence of strong homophily in academic achievements of high school and university students. We analyze a unique longitudinal dataset that contains information about the detailed evolution of a friendship network of 4.500 students across 42 months. Combining the evolving social network data with the time series of the academic performance (gpa) of the individual students, we show that academic homophily is a result of selection: students gradually reorganize their social networks according to their performance levels, rather than adapt their performance to the level of their local group. We are able to understand the underlying dynamics of grades and networks with a simple agent-based model. The lack of a social pull effect in classical educational settings could have important implications for the understanding of the observed persistence of segregation, inequality and limited social mobility in societies.

Molecular biology and microbiology have rendered a picture of living organisms as minutely orchestrated versatile machines and factories that are able to steer flow of matter with unprecedented control and accuracy. In my talk, I will present a portfolio of work that exploits the physical properties of biomolecules and super-molecular biological aggregates for non-biological, technological applications in the areas of molecular computing and material production. In the first part of the talk, I will present recent work on a DNA implementation of a signal recorder based on a stack data structure that is studied in vitro and in simulation. The second part of the talk expands on these principles and uses DNA computing to arrange and control the dynamics of vesicles that are able to carry molecular cargo. By combined molecular and external control mechanisms, we develop a framework to orchestrate bio-chemical synthesis pathways in a manner that mimics the organization of cellular compart-ments such as the Golgi apparatus. Time allowing, I will present some works that integrates the above approaches of refunctionalizing biomolecules to create life-like yet truly non-biological chemical aggregates de novo.

My work investigates the implications of systemic risk for financial regulation. I show that in the absence of liquidation costs for defaulted firms, there exists a social dilemma in which socially and pareto efficient bailouts are not performed in a Nash equilibrium by rational agents without government intervention. Next, I establish a general framework for measuring systemic risk capable of separating the effects of direct contagion, asset fire sales and mark-to-market effects. Using Austrian interbank data, I show that the importance of fire sale effects eclipses the effects of the other two channels. Banks-specific indicators of systemic importance are then constructed to assess the efficiency of Basel-III systemic risk regulations, which rely on a simple combination of bank-specific indicators and do not account for network effects. The empirical results suggest that the Basel-III indicator set is an efficient choice of bank-specific indicators, but also that such indicator sets generally perform poorly at capturing the importance of network effects for contagion.

Secondary use of linked administrative claims data in the healthcare and social insurance system unveils various new possibilities for research and decision making. The Main Association of Austrian Social Security Institutions (“Hauptverband der österreichischen Sozialversicherungsträger”, HVB) operates the so called “GAP-DRG” database, including reimbursement and pseudonymized personal data from the Austrian healthcare system as part of the K-Project DEXHELPP* at the TU Vienna. Two main topics concerning this data collection are addressed. First, current developments as well as the renewal of key components of the database infrastructure in the near future are discussed. Comprehensive changes of hardware and software in the last year – a completely new database server has been deployed – and in the following month are summarized. A novel application platform based on software containers, allowing centralized services and new ways to analyze and present data will be introduced besides improvements due to upgrades of the underlying hardware and software. Second, a concrete example of applied data analysis is presented, bringing together various segments of the database and details of the Austrian public health insurance system. Due to missing genealogical information in GAP-DRG, family relations have to be derived from co-insurance of dependents. Co-insured children and spouses can therefore be distinguished based on their age in relation to the insured person’s age. The presented algorithm can be fitted to diverse objectives concerning the age of parents, spouses and children, while quality control and visualization of multivariate coherences are emphasized. *http://dexhelpp.at

Diffusion of information, behavioral patterns or innovations follows diverse pathways depending on a number of conditions, including the structure of the underlying social network, the sensitivity to peer pressure and the influence of media. We introduce a general model that incorporates threshold mechanism capturing sensitivity to peer pressure, the effect of ‘immune’ nodes who never adopt, and a perpetual flow of external information and study it by analytical methods and simulations. While any constant, non-zero rate of dynamically-introduced spontaneous adopters leads to global spreading, the kinetics by which the asymptotic state is approached shows rich behavior. In particular we find that, as a function of the immune node density, there is a transition from fast to slow spreading governed by entirely different mechanisms. This transition happens below the percolation threshold of network fragmentation, and has its origin in the competition between cascading behavior induced by adopters and blocking due to immune nodes. This change is accompanied by a percolation transition of the induced clusters. We calibrate and validate the model using Big Data.

Many systems of processes can be characterized by sequences of symbol strings where symbols are sampled from an alphabet (or by sentences of words sampled from a lexicon). Moreover, such symbol sequences may be subject to internal constraints which determine which sequences a process can generate and which not. This can be interpreted as a „grammar'' that distinguishes well-formed sequences from ill-formed ones. For simple grammars (typically referred to as regular grammars) symbols in the underlying alphabet can be represented by matrices acting on a finite dimensional state-space. As a consequence it becomes possible to determine the multiplicity of sequences with identical histograms, and hence the maximum configuration entropy, i.e. the Boltzmann entropy, of the processes. We discuss the general situation in the particular example of the Oslo sandpile-model.

Petr Jizba and Fabio Scardigli In this talk I will show how a Brownian motion on a short scale can originate a relativistic motion on scales that larger than particle's Compton wavelength. I start by discussing complex dynamical systems whose statistical behavior can be explained in terms of a superposition of simpler underlying dynamics — the so-called superstatistics paradigm. Then I go on by showing that the combination of two cornerstones of contemporary physics — namely Einstein’s special relativity and quantum-mechanical dynamics is mathematically identical (when analytically continued to Euclidean regime) to a complex dynamical system described by two interlocked processes operating at different energy scales. The combined dynamic obeys special and doubly special relativity even though neither of the two underlying dynamics does. To model the double-stochastic process in question, I consider quantum mechanical dynamics in a background space consisting of a number of small crystal-like domains varying in size and composition, known as polycrystalline space (or Voronoi tessellation). There, particles exhibit a Brownian motion. The observed relativistic dynamics then comes solely from a particular grain distribution in the polycrystalline space. In the cosmological context such distribution might form during the early universe’s formation. Salient issues such as Hausdorff dimensions of path-integral trajectories, connection with Feynman chessboard model and implications for quantum field theory and cosmology (leptogenesis) will be also briefly discussed. Related articles: [1] P. Jizba and F. Scardigli, Eur. Phys. J. C (2013) 73: 2491 [2] P. Jizba and F. Scardigli, Phys. Rev. D (2012) 86: 025029

The internet and social network sites transformed the world. Instant access to the most of human knowledge and to the millions of peers registered in social networks means that people are not limited by their immediate environment anymore. Can this opportunity be used to break the cycle of inequality reproduction or will we observe the Matthew effect: the rich will get richer and the poor will get poorer? To shed light on this question I've collected a data set from the VK ("Russian Facebook") containing information about 1 million users, their social ties and interests along with their educational background. I'm going to use complex networks analysis in order to estimate the effect of segregation in virtual space and the role of peer influence in overcoming social background.

Quantitative social science, which strives for a predictive understanding of social facts, has since its origins faced the challenge of limited availability of empirical data. We approach this challenge by turning to a a model-society inside a large-scale virtual world, i.e. an online game, of which we have nearly complete information. Studying the structure of society as a whole, multiple levels of fractally nested groups and subgroups have been found in real-world anthropological data. By quantitatively analyzing detailed grouping data we find a similar structure in the model-society. This finding indicates that the pattern of fractal organization is universal. Humans within such a hierarchically organized society are clearly unequal. We contribute to the study of inequality by analyzing the wealth distribution inside the virtual world and by comparing it to real-world observations. Our data allow us to investigate connections between wealth and the position of individuals within their social networks. We study pairs of interacting individuals and quantify the interplay between physical distance and social interactions. Our findings are compared to previous real-world data.

The endeavor to grasp the meaning of text in Artificial Intelligence starts with understanding the natural building block of the language : „words“. Since years, many methods has provided various computer-understandable representations by trying to resemble the human perception of the meaning and relatedness of the words. The methods basically exploit either human-annotated knowledge or statistical approximation and are broadly used in different fields such as Natural Language Processing, Information Retrieval, Text Mining, Sentiment Analysis, Statistical Machine Translation, etc.. As a main direction for representing words, the statistical methods exploit the implicit knowledge within long text corpora to approximate the representation of words in vector spaces (Word Embeddings). These methods mainly rely on the tendency of natural language to use semantically related words in similar contexts. In the text retrieval community, word and text representation started with Latent Semantic Analysis/Indexing (LSA/LSI), the pioneer approach that initiated a new trend in surface text analysis. Explicit Semantic Analysis (ESA) is one of the early alternatives, aimed at reducing the computational load. However, unlike LSA, ESA does rely on a pre-existing set of concepts, which may not always be available. Random Indexing (RI) is another alternative to LSA/LSI that creates context vectors based on the occurrence of words contexts. It has the benefit of being incremental and operating with significantly less resources while producing similar inductive results as LSA/LSI and not relying on any pre-existing knowledge. Word2Vec further expands this approach using neural networks-based language modeling. When trained on large datasets, it is also possible to capture many linguistic subtleties (e.g., similar relation between Italy and Rome in comparison to France and Paris) that allow basic arithmetic operations within the model. In this talk, we first review the main word representation methods with focus on Word2Vec as an effective and state-of-the-art approach. Putting forward the question of the meaningfulness of the statistical word representations, we then shed light on the topology of the distributional space in different dimensions, based on recent studies in the IFS group of TU WIEN. The analysis provides an understanding about the informativeness versus randomness of the words' relations and motivates the effective use of the word embeddings in different areas.

We review a particular approach to generalized entropy formulas based on finite reservoir physics. Ideal systems - correlated only due to restrictions of the total phase space - reveal certain leading order corrections to the traditional entropy formula first suggested by Boltzmann and later extended in use and interpretation by Gibbs, Planck and Shannon. We identify which axioms behind the classical formula are violated by this finiteness and how the large dimension scaling may lead to further formulas beyond the well-studied logarithm.

Different from traditional quantitative linguistics, we study languages in the following ways: distribution of the appearing frequencies in the text, emerging speed of new words, nested structure of the potential followers and information needed for creating word sequences among other approaches. Each of the above analyses tells us some properties, either from word-word level or the text as a whole. Usually these properties are studied for a book or a specific language, while other detailed aspects remain vacuum. Thanks to the Corpus of Historical American English(COHA), we get the about more than 110,000 texts ranging from 1810 to 2009 with four genres of fictions, magazines, newspaper articles and non-fictions. The texts are analyzed using the above approaches across history. From the results we observe clear trends in these properties, which indicates that the English language is evolving. Also, texts are decomposed in sentences and then sentences of the same length are clustered and then analyzed. We observe patterns that stay invariant through history. This may indicate the invariance of the habits of language use. The part-of-speech analysis for different sentence length are also presented for the possible explanation of this invariance. These findings may lead to a more comprehensive understand of the statistical properties of language.

The comprehension of the mechanisms behind scaling patterns has become on of the hot topics of modern statistical physics [1]. From complex networks to critical phenomena, scaling laws emerge in somewhat regular way. In this talk I will link the scaling patterns observed in many complex systems with a crucial property behind them: history-dependence [2]. Classical examples of history dependent processes with extremely interesting properties are Pólya urns, the Chinese restaurant or the recurrent random sequences proposed by Ulam and Kac. The link between history-dependence and scaling comes from the recently defined Sample Space Reducing (SSR) processes [3,4]. SSR processes connect history dependence and scaling in an extremely intuitive way. In addition, SSR process are a totally new route to scaling which can explain a huge range of power-law exponents thanks to the unique assumption that the sampling space is reduced as long as the process unfolds. Simple forms of SSR processes are regular sampling processes where the ‘left-right’ symmetry is broken, leading to a minimal form of history-dependence. In spite of the simplicity of this basic assumption, SSR processes display a wide spectrum of surprising properties. From this zoo of interesting properties, maybe the most remarkable one is the role of the scaling law known as ‘Zipf’s law’ as an attractor, which provides a new, fundamental explanation for the ubiquity of such scaling pattern in real systems. In addition, it provides a new look to diffusive phenomena, since it assumes the presence of a 'target' or 'sink' region, which may model many diffusion-like phenomena, such as urban movement, animal migration or information routing in the internet. The intuitive rationale behind the SSR processes and the surprisingly simple mathematical apparatus needed to understand them makes the SSR process approach a new research area with promising applications. Bibliography: [1] Newman, M E J (2005) “Power laws, Pareto distributions and Zipf's law" Contemporary Physics 46 (5): 323–351. [2] Arthur, B (1994) Increasing Returns and Path Dependence in the Economy. The University of Michigan Press, Ann Arbor. [3] Corominas-Murtra, B, Hanel, R, and Thurner, S (2015) “Understanding scaling through history-dependent processes with collapsing sample space”. Proc Nac Acad Sci USA. 112 (17) 5348–5353. [4] Thurner, S, Hanel, R, Liu, B, and Corominas-Murtra, B (2015) “Understanding Zipf’s law of word frequencies through sample-space collapse in sentence formation”. J R Soc Interface. 12 (108).

The fundament of a health democracy are free and fair elections. Election forensics is an emerging, interdisciplinary field that aims to develop statistical and data-driven methods to verify whether the preferences of the electorate have been correctly translated into the election outcome. One of the most often reported types of election fraud is voter coercion. This is the practice of intimidating, threatening, or coercing any person for voting or for attempting to vote. Here, we develop a novel statistical method to identify polling centers that show substantial deviations in their electoral behavior with respect to other, geographically close polling centers. We analyse a dataset of twenty different elections in ten different countries and show that in the vast majority of cases no such systematic deviations exist, with the exceptions of electoral referendums after 2003 in Russia, after 2006 in Venezuela, and, to a smaller extent, after 2012 in Mexico. In each of these cases we find a substantial number of polling centers with significantly increased results of turnout and votes for the winning party. Polling centers with a comparably small electorate size are particularly susceptible to such deviations that are compatible with the assumption of widespread voter coercion. We discuss these findings in the context of recent elections in Venezuela, where the deviations are strongest in newly created and strategically located polling centers.

Human societies can be viewed as complex systems comprising individuals that are simultaneously active in various social networks, each of the latter representing a different type of social relation. Examples for such relations are friendship, enmity, communication, or trade. Dynamical changes of a social system's overall structure due to interactions among individuals are accordingly reflected in structural changes of the corresponding social networks. In the past, some of the observed structural properties of single social networks (e.g. friendship network) could be successfully explained by mathematical models, in the sense that those models generate artificial networks with similar properties. But in general, the individual networks of a social system are not independent from each other: According to structural balance theory, certain triangular configurations of positively connoted relations (such as friendship) and negatively connoted relations (such as enmity) are more balanced that others, and thus more likely to occur. Using a multilayer network formalism, we are currently developing a mathematical model for the dynamics of social systems with one friendship and one enmity network layer. The model is based on local dynamical rules that implement the key processes preferential attachment and triadic closure, and especially respect structural balance theory.

Self reinforcing processes are ubiquitously at work in complex dynamical systems in terms of feed-back loops. Polya urns provide a general framework to study the statistics of such processes and their distribution functions. Maximizing entropy under particular conditions (the maximum entropy principle) allows to predict how likely particular types of systems (weakly interacting components, equilibrium conditions, Markovian, etc.) can be found in particular states. The classical MaxEnt approach however breaks down for non-equilibrium processes such as Polya urn processes. However, using standard maximum configuration considerations we can derive the adequate MaxEnt principle for multi-state Polya urn processes. The form of entropy and constraints to consider in MaxEnt are direct consequences of the class of processes MaxEnt is supposed to apply. This observation has far reaching consequences for an information theory of self reinforcing processes.

Innovation is the driving force in the evolution of human society as well as of biological systems. A general concept that applies to innovation and the emergence of novelties, is what Kauffman called expanding the adjacent possible. The idea is that by creating fresh opportunities, one novelty can pave the way for others, enlarging the space of possibilities in a self-consistent way. I will highlight some statistical features common to many systems where some kind of innovation occurs, and I will shortly review seminal works devoted to their comprehension. I will then present a recent work aimed at grounding the notion of adjacent possible on real data, by the definition of quantitative measures and the development of a suitable mathematical framework.

Power laws appear in a wide variety of physical, biological and man-made phenomena. They can also be observed in frequencies of human diseases. Over a wide range of magnitudes, rank ordered disease frequencies follow an approximate power-law. Several mechanisms have been proposed to understand the origin of power-law behavior of complex systems. A recently proposed approach offers an alternative way to understand scaling, based on sample space reducing (SSR) processes. In respect to the topology of directed networks, SSR can be related to diffusion processes. It was shown that progression of human diseases can be explained by diffusion on phenotypic disease networks (PDN). I will talk about our studies of diffusion-processes on PDN's and their relation to SSR. I will also discuss if we are able to understand the emergence of scaling in disease frequencies by using a SSR approach.

With the advent of Synthetic Biology, methods for the computational design of catalysed reaction networks became important. I will present a computational framework, which combines a graph-grammar base formalism for describing chemical transformations with optimal flows on hypergraphs to find solutions for the design problem. An optimal network is perceived from a set of initial molecules and the set of allowed enzymatic reactions.

Classical microbiology has focused on single species. Yet in natural environments and in many industrial applications microbes exist in functional consortia or communities (MCs). MCs play crucial roles in processes as divers as global climate regulation, human health or industrial applications such as wastewater treatment. While sequence-based studies in the last decade revealed immense biodiversity and complexity, we are far from understanding the structure and organisation of MCs and how these translate to function and productivity of microorganisms. Co-occurrence networks of microorganisms from environmental samples facilitate understanding and analysis of this structure-function relationship. In my talk I will focus exemplarily on one ecological and one pathological MCs and their organisation. I will show how the analysis of MC network topology allows inference of functional roles in riverine biofilm MCs and demonstrate that environmental perturbation leads to network fragmentation. For the pathologic MC (cystic fibrosis lung microbiota) we could confirm by network analysis that a shift in the severeness of the condition is related to a taxonomic shift in the MC and the related metabolic core processes. Furthermore we propose functional and taxonomical keystones as targets for novel drug development.

Social balance theory suggests that within a social network, the subnetworks comprising positive ties (friendship network) respectively negative ties (enmity network) are dynamically interrelated. I will present some results from my recent work which concentrates on finding a realistic generative model for such networks based on data from the massive multiplayer online game Pardus.

Each sphere of knowledge and information could be depicted as a complex mesh of correlated items. By properly exploiting these connections, innovative and more efficient navigation strategies could be defined, possibly leading to a faster learning process and an enduring retention of information. It has been investigated how the topological structure embedding the items to be learned can affect the efficiency of the learning dynamics. To this end we introduce a general class of algorithms that simulate the exploration of knowledge/information networks standing on well-established findings on educational scheduling, namely the spacing and lag effects. While constructing their learning schedules, individuals move along connections, periodically revisiting some concepts, and sometimes jumping on very distant ones. In order to investigate the effect of networked information structures on the proposed learning dynamics we focused both on synthetic and real-world graphs such as subsections of Wikipedia and word-association graphs. We highlight the existence of optimal topological structures for the simulated learning dynamics whose efficiency is affected by the balance between hubs and the least connected items. Interestingly, the real-world graphs we considered lead naturally to almost optimal learning performances.

The patterns of joint activity in a population of retinal ganglion cells encode the complete information about the visual world, and thus place limits on what could be learned about the environment by the brain. We analyze the recorded simultaneous activity of more than a hundred such neurons from an interacting population responding to naturalistic stimuli, at the single spike level, by constructing accurate maximum entropy models for the distribution of network activity states. This – essentially an ”inverse spin glass” – construction reveals strong frustration in the pairwise couplings between the neurons that results in a rugged energy landscape with many local extrema; strong collective interactions in subgroups of neurons despite weak individual pairwise correlations; and a joint distribution of activity that has an extremely wide dynamic range characterized by a zipf-like power law, strong deviations from ”typicality”, and a number of signatures of critical behavior. We hypothesize that this tuning to a critical operating point might be a dynamic property of the system and suggest experiments to test this hypothesis.

Cell phones are now ubiquitous: it is estimated that the number of phones in use exceeds the size of the global population in 2015. I will talk about two main lines of work in our lab. First, we use call detail records (CDRs) to investigate the structure of large-scale social networks and their relationship to underlying geography. We also use these data to study population dynamics at massive gatherings of people and to inform models of epidemic spread, in particular cluster randomized trials. Second, we have coined the term digital phenotyping to refer to the moment-by-moment quantification of the individual-level human phenotype, in situ, using data from digital devices. Using a dedicated smartphone application, we collect active data (surveys, voice samples, etc.) and passive data (spatial location, social engagement, etc.) from consenting patients and analyze these data using modern statistical learning methods. I will talk about some of our work that utilizes these two approaches, what types of insights they may yield to the study of social networks and human behavior, and how this work interfaces with public health.

Recent advances in disease gene identification and high-throughput mapping of physical interactions between gene products offer new opportunities to explore the role of molecular networks in human disease. Here we show that proteins associated with the same disease display a statistically significant tendency to agglomerate in the same neighborhood of the interactome, offering quantitative evidence for the existence of well-localized and potentially identifiable disease modules. Most important, we find that the network-based location of each disease module determines its pathobiological relationship to other diseases. For example, disease pairs with overlapping modules show significant co-expression patterns, symptom similarity, and comorbidity; those that reside in separated network neighborhoods are pathobiologically and clinically distinct. The proposed interactome-based framework offers systematic avenues to discover common molecular roots between clinically unrelated disease phenotypes even if they do not share disease genes, and helps identify the biological role of GWAS genes of small effect size and low genome-wide significance.

This talk will briefly review the history of risk measures appearing in the subsequent generations of international banking regulation, their relative merits and shortcomings, with special emphasis on Expected Shortfall (ES) that is on its way to becoming the next global regulatory market risk measure. When these measures are used to predict the out-of-sample risk or to optimize portfolios they all display a weakness due to the relative scarcity of data compared to the size of institutional portfolios. The resulting estimation errors can be very large, in fact, for some critical values of the parameters they can diverge, with the optimization algorithm undergoing a phase transition. The estimation error problem is particularly serious for downside risk measures, such as ES. The standard way to tackle these high dimensional problems is to use regularization, that is to introduce a penalty for excessive positions. It is shown that investors who optimize their portfolios under any of the coherent risk measures are naturally led to regularized portfolio optimization when they take into account the impact their trades make on the market. The impact function determines which regularizer is to be used. It is shown that any regularizer based on the norm Lp with p > 1 makes the sensitivity of coherent risk measures to estimation error disappear, while regularizers with p < 1 do not. The L1 norm represents a border case: its “soft” implementation does not remove the instability, but rather shifts its locus, whereas its “hard” implementation (equivalent to a ban on short selling) eliminates it. We demonstrate these effects on the important special case of Expected Shortfall.

Lindley's paradoxon is an effect that haunts Bayesian Hypotheses testing especially if hypotheses such as H1: A and B are correlated, versus H0: A and B are independent, are tested against each other. There is of course no real paradoxon only prior probabilities that have been chosen badly. We show that by a simple technique, comparing extreme samples which represent the hypotheses in their purest obtainable form, allows to gage the prior probabilities of the hypotheses adequately. We analyze the situation in the context of binary processes and show that in the limit of large numbers of samples the posterior probabilities of the hypotheses H1 and H0 exhibit a first order phase transition at critical values of the correlation coefficient.

The number of research papers published yearly shows a staggering exponential growth. One can in fact witness a century-long publication avalanche. According to the two major databases INSPEC and Thomson Reuters, there is a 300-fold increase in the number of published items since the year 1900. The world of science has recently reached such a level of proliferating complexity that its structural evolution itself poses challenging scientific problems.

Novel colloidal systems constitute a broad and developing new field of research in soft matter science. Telechelic star polymers (TSPs), which are a hybrid system that combines the flexibility of chains with the sphericity of colloids, are particularly interesting materials because of their high degrees of freedom and the complex structural and dynamical properties that arise from them. This talk will focus on the ability of TSPs to self-assemble, to undergo macroscopic as well as microscopic phase transitions and to form multiply-connected percolating networks, depending on their functionality and the ratio of solvophilic to solvophobic groups.

The concept of networking is used in many applications: computational processes (e.g., neural networks), transport, ecology, economics, sociology and many others. Therefore, understanding the dynamics of large, complex networks is becoming increasingly important. In these systems, stability is of utmost importance. Instabilities always lead to disaster and can even assume global proportions. Whereas stable systems can regenerate themselves and thus have some degree of fault tolerance, making them partially immune even to attacks. The question of the conditions for stability in networks is a central theme of my lecture.

Phase transitions are one of the most interesting topics in physics. They occur in various physics-related disciplines from thermodynamics to network theory. In my talk I'd like to present the content of the paper "Phase transition in random catalytic networks" (by Rudof Hanel, Stuart A. Kauffman and Stefan Thurner). I'll start with some usefull definitions for the description of catalytic networks (like the "product rule density", the "support of a process", the "forward difference", etc.). Afterwards I'll derive the growth equation for catalytic sets and an analytical approximation of the forward-closure size. Plotting the "final number of products" against the "pair density" will lead to diagrams equal to PV-diagrams when considering Van der Waals equation and therefore we'll identify phase transition in catalytic networks.

Phase transitions are ubiquitous in Nature. Interestingly, even spacetime itself can be subject to such transitions, like in the famous Hawking-Page phase transition between 'hot curved space' and certain black holes. Recently, we found a similar phase transition between 'hot flat space' and an expanding cosmological spacetime in 2+1 dimensions. I review the ingredients contained in our 'cooking recipe' for a Universe.

The notion of phase transitions is frequently used as a metaphor for abrupt changes in evolutionary dynamics. Here it will by understood more rigorously in the context of selection mutation dynamics. Replication and mutation are modeled as parallel chemical reactions, and evolutionary dynamics is seen as an adaptive or a random walk of population in an abstract space of genotypes represented by nucleic acid sequences. Depending on the distribution of fitness values over sequence space, populations can tolerate replication errors only up to a certain maximal mutation rate, which has been characterized as an error threshold. Below threshold and in the long time limit the populations converge to stationary sequence distributions denoted as quasispecies, whereas above threshold the populations are permanently non-stationary and drift randomly through sequence space. Accordingly, error thresholds are transitions from localized quasispecies to migrating populations, and they become sharper and sharper with increasing lengths of sequences. They have much in common with phase transitions. In the seminar we shall explore the conditions for the existence of error thresholds and analyze the nature of the transitions.

I first describe simple statistical physics models discribing phase transitions in various dimensions. Ferromagnetic ones are related to lattice regularized Euclidean field theory. In dimensions greater than four one obtains triviality. Next I describe renormalization group ideas and mention the Landau ghost problem. An improvement is obtained by deforming space-time. Together with Raimar Wulkenhaar we found a model which is asymptotic safe and which was constructed recently. The connection to matrix models is mentioned.

In this overview, we start from the notion of a phase and ask ourselves why it is important to understand and distinguish phases, which leads us to the high cultural and technological significance of multi-phase structures. The equilibrium states of phases such as displayed in phase diagrams can be sought by means of thermodynamic potentials. We discuss how to calculate the quantities found therein. Regarding two-component (as the simplest case of multi-component) systems, it is seen how phase diagrams can easily be derived by graphical thermodynamics. Thermodynamics requires that on entering a miscibility gap a new phase be generated. This leads us to precipitation as a particularly interesting case of phase transformation, with nucleation and spinodal decomposition as alternative pathways. Opposed, in a certain sense, to phase separation are ordering transitions. Atomic configuration can be described by various order parameters based on correlation functions. We consider short-range order and long-range order. The Bragg-Williams model of ordering is the simplest ansatz leading to meaningful qualitative behavior. We recognize in it a one-point mean field approximation to the Ising model. In the framework of the Ising model ordering / phase- separating alloys can be seen analogous to magnetic systems. Finally, martensitic phase transformations are a typical solid-state phenomenon. Once initiated, they proceed almost instantaneously, practically at the speed of sound, and allow remarkable applications such as shape-memory alloys.

Assume a seller wants to sell a digital product in a social network where nodes represent clients and edges represent a friend-like relationship between them. Assume further that the value of the item for a buyer has positive network externalities, i.e., it is a monotone function of the set of neighbors that already have the item. The goal of the seller is to maximize his revenue. Previous work on this problem studies the case where clients are offered the item in sequence and have to pay personalized prices. This is highly infeasible in large scale networks such as the Facebook graph: (1) Offering items to the clients one after the other consumes a large amount of time, and (2) price-discrimination of clients could appear unfair to them and result in negative client reaction or could conflict with legal requirements. On the other side, offering the same price to every client significantly reduces the seller’s revenue. Thus, we study settings where we limit the duration of the selling process and the amount of price discrimination. Specifically, the item is offered in parallel to multiple clients at the same time and at the same price. This is called a round. We show that with O(log n) rounds, where n is the number of clients, a constant factor of the revenue with price discrimination can be achieved and that this is not possible with o(log n) rounds. Moreover we show that it is APX-hard to maximize the revenue even if the externalities in the clients’ valuation functions are very restricted, and we give constant factor approximation algorithms for various further settings of limited price discrimination.

A network representation is useful for describing the structure of a large variety of complexsystems. However, most real and engineered systems have multiple subsystems and layers of connectivity,and the data produced by such systems is very rich. Achieving a deep understandingof such systems necessitates generalizing \traditional" network theory, and the newfound deluge of data now makes it possible to test increasingly general frameworks for the study of networks.In particular, although adjacency matrices are useful to describe traditional single-layer networks,such a representation is insucient for the analysis and description of multiplex and time-dependent networks. One must therefore develop a more general mathematical framework to cope with the challenges posed by multi-layer complex systems. In this paper, we introduce a tensorial framework to study multi-layer networks, and we discuss the generalization of several important network descriptors and dynamical processes|including degree centrality, clustering coecients, eigenvector centrality, modularity, Von Neumann entropy, and diusion|for this framework.

What is the core of a complex network? In this talk we will discuss this crucial question of modern complex networks theory. To start with, we will briefly revise the underlying theory of network decomposition and nested subgraphs. We will then discuss specific core-finding and network decomposition methods. Special emphasis will be given to the K-core, the paradigmatic method of core decomposition. It is worth to note that the K-core is nowadays a source of both mathematical and physical problems, but that it was proposed more than 30 years ago to disentangle the deep structure of social networks. However, this core decomposition mechanism has several limitations that lead to new, more sophisticated methods, like the GK-core or the M-core, which better agree to the notion of core in specific problems, specially those related to social networks. We will finally revise some applications of this formal/algorithmic machinery to real problems, like the understanding of the network dynamics defined by virtual society of the massive on line game 'Pardus'. [1] Seidman, Stephen B. (1983), "Network structure and minimum degree", Social Networks 5 (3): 269–287 [2] Dorogovtsev, S.N.; Goltsev, J.F.F.; Mendes, JF (2006), "k-core organization of complex networks", Physical Review Letters 96 (4): 040601 [3] B Corominas-Murtra, JFF Mendes, RV Solé (2008) "Nested subgraphs of complex networks", Journal of Physics A: Mathematical and Theoretical 41 (38), 385003 [4] Bernat Corominas-Murtra, Benedikt Fuchs, Stefan thurner "B Corominas-Murtra, B Fuchs, S Thurner" arXiv preprint http://arxiv.org/abs/1309.6740 [5] Pol Colomer-de-Simón, M. Ángeles Serrano, Mariano G. Beiró, J. Ignacio Alvarez-Hamelin, Marián Boguñá "Deciphering the global organization of clustering in real complex networks", Scientific Reports 3, Article number: 2517 doi:10.1038/srep02517

I will give an detailed introduction to Markov theory in quantum and classical contexts and its relation to the "measurement problem" in quantum mechanics. This is followed by an introduction to some notions in graph theory and random walks. The talk will then centered on our article "Quantum Stochastic Walks" Phys. Rev. A, vol 81 page 022323 (2010) and the axiomatic introduction of the quantum stochastic walk. We will then discuss related results on thermodynamics, directed quantum transport, and Lieb-Robinson bounds.

Statistical mechanics of many statistical systems allows to asymptotically replace expectation values by the so called `maximum configuration'. Sampling processes, where the probabilities of observing a system in particular states evolves stationary, can be described by the reduced Boltzmann entropy that in such cases takes the functional form of Shannon entropy. The probability of observing a non Markovian, aging system in a certain state is by definition not a stationary process and Boltzmann entropy of such systems becomes non extensive. The `maximum configuration' argument on the other hand still holds also for sufficiently large aging systems. Using that entropy is a function that only depends on the frequencies of observing the system in a particular state allows to show that the reduced Boltzmann entropy of aging systems takes the functional form of generalized entropies, as for instance Tsallis entropy.

Barnett et al. in 2009 proved that Granger causality and transfer entropy causality measure are equivalent for time series which have a Gaussian distribution. Granger causality test is linear, while transfer entropy a non-linear test. Many biological and physical mechanisms show to have non-Gaussian distributions. In this paper we investigate under which conditions on probability density distributions of the data can the equivalence of the two causality measures be extended. In the complexity sense ”cheaper” linear Granger test can be applied for detection of causality in time series satisfying these conditions. These results have an impact on causality detection in common biological and physical time series.

We use a toy model of the financial market to test the efficiency and dangers of credit regulation schemes. We find that Basle-type regulation works fine in situations of low leverage levels in the financial system, however they become destabilizing in scenarios with realistic leverage level. We further design an ideal world, where all leverage introduced risk is hedged with options. Even by assuming that option writers never default, we see that introducing the heavy requirement of complete hedging does not make the system systemically more secure.

In the last decade increasing attention has been devoted to the study of random walks on complex topologies. Various features of random walks on networks, such as passage times and spectral properties have been investigated, and random walks have also been used to to detect communities, to evaluate centrality of nodes and to coarse-grain graphs. Another quantity recently considered for the study of random walks is the entropy rate, a measure used to characterize the mixing properties of a stochastic process. In this talk we will first discuss some of the properties of random walks which might have many relevant applications. In particular, we will consider biased random walks, i.e. random walks with a jumping probability which depends on some properties of target node. We will then focus on designing biased random walks with maximal entropy rate on a given graph, i.e. choosing the transition probabilities of the random walk in such a way that the random walkers are maximally dispersing in the graph, exploring every possible walk with equal probability. Although in principle the optimization of the entropy rate requires that a walker has, at each time step, a global information on the structure of the entire graph, information which is often unavailable, we will demonstrate that it is always possible to construct maximal-entropy random walks by defining a set of transition probabilities that are markovian and that rely only local information on the graph structure.

Zentrale Teile der Stochastik kreisen um das Gesetz der großen Zahlen, wonach (bereits unter schwachen Voraussetzungen) die Konvergenz von Mittelwerten gegen den Erwartungswert fast sicher ist. Aus topologischer Sicht, d.h. im Sinne Bairescher Kategorien, ist jedoch ein gänzlich konträres Verhalten typisch. Der allgemeine mathematische Hintergrund dieses Phänomens soll ebenso beleuchtet werden wie einige charakteristische Beispiele.

A short introduction to the concept of generalized entropies: The necessity of generalized entropies is not primarily grounded in alternative means to reproduce particular types of observed distribution functions from some modified maximum entropy principle - but by the concept of Extensivity. By relaxing the axioms of Information Theory an axiomatic approach towards generalized entropies can be taken. The generic functional form of generalized entropies can be deduced as well as conditions which allow to determine which entropy will be the extensive entropy of some system - given the knowledge of how phase-space expands as a system is enlarged.

The regulation of gene expression in live-forms is subject to high > dynamics and nonlinearities. > To create a mathematical model of these systems biological processes > must be reflected adequately. > This talk will handle the biological foundations of gene expression from > DNA to RNA including eukaryotic characteristics like RNA splicing with > respect to the influence on the mathematical model.

We study long-time mobility of human-controlled characters on a network-shaped universe of a massive multiplayer online game. We take a number of mobility measurements and compare them with measures of simulated random walkers on the same topology. Mobility of players is sub-diffusive - the mean squared displacement follows a power law with exponent 0.4 - and significantly deviates from mobility patterns of random walkers. Mean first passage times and transition counts relate via a power-law with slope -1/3. We compare our results with studies where human mobility was measured via mobile phone data and find striking similarities.

Explaining the breadth of human cooperation is a prime challenge in both the social sciences and biology. One possible mechanism focuses on the role of reputation in cooperation. Usually it is assumed that such a reputation is objective--that is, the same for all agents. We develop a model where reputations are private and synchronized through the sharing of information, i.e. gossiping. Interpreting cooperation between two agents as a positive link and defection as a negative link, this model shows an interesting connection to social balance theory. Furthermore, we study its evolutionary stability.

Strongly interacting statistical systems – complex systems in particular – can change their macro- scopic properties merely as a function of the number of their constituents. Examples include neurons, state-forming insects, financial markets, etc. where systemic properties of small systems can differ drastically from those of a large system built from the same components. The origin of this property is not understood on fundamental grounds. Here we explore this phenomenon from first principles within a thermodynamical framework, by asking about the consequences of bringing interacting sub- systems in thermal contact, where the first three Kinchin axioms hold but the 4th is violated. We show that all sufficiently interacting statistical systems fall into two categories: systems which are asymptotically stable, and those which are asymptotically unstable, meaning that small changes in system size can lead to a drastic increase in entropy. We argue that complex systems belong to this unstable class which make drastic qualitative changes possible as a function of system size. Under the same conditions we then derive the unique asymptotic entropy, Scd = Gamma (d + 1,1 - c ln pi) (c, d constants) which covers all equivalence classes of asymptotically stable and unstable, i.e. all interacting and non-interacting systems. The corresponding distribution functions are special forms of Lambert-W exponentials. As special cases they contain Boltzmann, stretched exponential and Tsallis distributions (power-laws) – all widely abundant in nature. This is, to our knowledge, the first ab initio justification for the existence of generalized entropies.

The celebrated statistical mechanics introduced by Boltzmann and Gibbs more than one century ago lie (for classical systems, for instance) on hypotheses such as ergodicity and mixing. Strongly chaotic systems, with positive maximum Lyapunov exponent, satisfy requirements of this sort. Within this realm, relevant random variables are probabilistically independent or nearly so. It is for such situations, and related quantum ones, that the central limit theorem and the standard entropy (Boltzmann, Gibbs, von Neumann, Shannon) exhibit their well known utility and connections with classical thermodynamics. What can be done outside this world? Can we approach such anomalous, and nevertheless ubiquitous, cases on thermostatistical grounds similar to the usual ones? For wide classes of such systems the answer appears to be positive, by appropriately generalizing the entropy and, consistently, the central limit theorem. Some central concepts as well as typical verifications and applications for natural, artificial and social systems will be briefly presented. BIBLIOGRAPHY: (i) C. Tsallis, Entropy, in Encyclopedia of Complexity and Systems Science (Springer, Berlin, 2009); (ii) C. Tsallis, Introduction to Nonextensive Statistical Mechanics - Approaching a Complex World (Springer, New York, 2009); (iii) S. Umarov, C. Tsallis, M. Gell-Mann and S. Steinberg, J. Math. Phys. 51, 033502 (2010); (iv) CMS Collaboration, J. High Energy Phys. 02, 041 (2010); (v) http://tsallis.cat.cbpf.br/biblio.htm

Human societies can be regarded as large numbers of locally interacting agents, connected by a broad range of social and economic relationships. These relational ties represent e.g. the feeling a person has for another (friendship, enmity, love), communication, exchange of goods, or behavioral interactions. Each type of relation spans a social network on its own. A whole society can be understood systemically only by uncovering interactions between different networks. Here we analyze a complete, multi-relational social network of a society consiting of over 300,000 players of a massive multiplayer online game. We extract networks of six different types of one-to-one interactions between the players. Three of them carry a positive connotation (friendship, communication, trade), three a negative (enmity, armed aggression, punishment). We first analyze these types of networks as separate entities and find that negative interactions differ from positive interactions by their lower reciprocity, weaker clustering and fatter-tail degree distribution. We then proceed to explore how the inter-dependence of different network types determines the organization of the social system. In particular we study correlations and overlap between different types of links and demonstrate the tendency of individuals to play different roles in different networks. As a demonstration of the power of the approach we present the first empirical large-scale verification of the long-standing structural balance theory, by focusing on the specific multiplex network of friendship and enmity relations.

Quantification of human group-behavior has so far defied an empirical, falsifiable approach. This is due to tremendous difficulties in data acquisition of social systems. Massive multiplayer online games (MMOG) provide a fascinating new way of observing hundreds of thousands of simultaneously socially interacting individuals engaged in virtual economic activities. We have compiled a data set consisting of practically all actions of all players over a period of three years from a MMOG played by 300,000 people. This large-scale data set of a socio-economic unit contains all social and economic data from a single and coherent source. Players have to generate a virtual income through economic activities to 'survive' and are typically engaged in a multitude of social activities offered within the game. Our analysis of high-frequency log files focuses on three types of social networks, and tests a series of social-dynamics hypotheses. In particular we study the structure and dynamics of friend-, enemy- and communication networks. We find striking differences in topological structure between positive (friend) and negative (enemy) tie networks. All networks confirm the recently observed phenomenon of network densification. We propose two approximate social laws in communication networks, the first expressing betweenness centrality as the inverse square of the overlap, the second relating communication strength to the cube of the overlap. These empirical laws provide strong quantitative evidence for the Weak ties hypothesis of Granovetter. Further, the analysis of triad significance profiles validates well-established assertions from social balance theory. We find overrepresentation (underrepresentation) of complete (incomplete) triads in networks of positive ties, and vice versa for networks of negative ties. Empirical transition probabilities between triad classes provide evidence for triadic closure with extraordinarily high precision. For the first time we provide empirical results for large-scale networks of negative social ties. Whenever possible we compare our findings with data from non-virtual human groups and provide further evidence that online game communities serve as a valid model for a wide class of human societies. With this setup we demonstrate the feasibility for establishing a 'socio-economic laboratory' which allows to operate at levels of precision approaching those of the natural sciences.

We present a general method to convert an ensemble of symbolic sequences into a weighted directed network. The nodes of the network are short motifs of the ensemble, while the directed links and their weights are defined from statistically significant co-occurrences of two motifs in the same sequence. The method is shown to be able in correlating sequences with functions in protein data, and might find useful applications for structure discovery as well as in computational linguistic and in the theory of dynamical systems.