Leiden Complex Networks Network (LCN2)
LCN2 organizes seminars on the last Friday of each month.
Previous LCN2 seminars will be listed here. As a guideline, each seminar will take place on the last Friday afternoon of every month.
1st seminar (March 27, 2015)
Breaking of ensemble equivalence in complex networks (slides)
Prof. dr. Frank den Hollander
It is generally believed that, for physical systems in the thermodynamic limit, the microcanonical description as a function of energy coincides with the canonical description as a function of temperature. However, various examples have been identified for which the microcanonical and canonical ensembles are not equivalent. A complete theory of this intriguing phenomenon is still missing.
In this talk we show that ensemble nonequivalence can manifest itself also in discrete enumeration problems. As specific examples, we consider ensembles of graphs with topological constraints. We show that, while graphs with a given number of links are ensemble-equivalent, graphs with a given degree sequence are not. This mathematical result provides a theoretical explanation for various 'anomalies' that have recently been observed in networks.
While it is generally believed that ensemble nonequivalence is associated with long-range interactions, our findings show that it may naturally arise in systems with local constraints as well.
Joint work with Diego Garlaschelli, Joey de Mol and Tiziano Squartini
2nd seminar (April 24, 2015):
Financial networks, systemic risk, and early-warning signals (slides)
Dr. Diego Garlaschelli
The financial crisis shifted the interest from traditional measures of "risk" of individual banks to new measures of "systemic risk", defined as the risk of collapse of an entire interbank system. Estimating systemic risk requires the knowledge of the whole network of exposures among banks. However, due to confidentiality issues, banks only disclose their total exposure towards the aggregate of all other banks, rather than their individual exposures towards each bank. Is it possible to statistically reconstruct the hidden structure of a network in such a way that privacy is protected, but at the same time higher-order properties are correctly predicted? In this talk, I will present general network reconstruction methods and discuss their remarkable performance on various economic, social, and biological networks. Then, as a counter-example, I will show an analysis of the Dutch interbank network done in collaboration with the Dutch Central Bank. We found that many standard topological properties of this network (such as the number of pairs of banks with mutual connections) display an abrupt change in 2008, providing a clear - but unpredictable - signature of the crisis. By contrast, after controlling for the heterogeneous connectivity of banks, the same properties show a gradual transition to the crisis, starting in 2005 and preceded by an even earlier period during which anomalous debt loops could have led to the underestimation of counter-party risk. Unlike the other examples discussed in the talk, these early-warning signals are undetectable if the network is reconstructed from partial bank-specific data. We discuss important implications for network modelling and bank regulatory policies.
3rd seminar (May 29, 2015):
Networks of the biological clock
Prof. dr. Johanna H. Meijer
Proper network-theoretical models of the brain are in need of experimentally well-understood brain areas that can be preferably studied at a multi-scale level. The suprachiasmatic nucleus (SCN) is the master clock in the mammalian brain and consists of 20,000 individually oscillating cells. Each cell contains a molecular feedback loop that produces an endogenous rhythm with its own intrinsic frequency. In order to obtain a robust and coherent 24-h rhythm that can drive other circadian rhythms in our body, the SCN cells synchronize to each other as a result of neural coupling. In addition to the internal synchronization, the SCN synchronizes to external cycles, such as to the 24-h light-dark cycle and to seasonal cycles. The network structure of the SCN results in a system that shows a balance between robustness on the one hand and flexibility on the other hand.
In our lab we perform electrophysiological recordings from single neurons and from populations of about 100 neurons. Furthermore, transgenic luciferase expressing mice are used to simultaneously measure the rhythms in gene expression at single cell level. Finally we record electrical activity from populations of neurons with implanted electrodes in freely moving animals. In this preparation, the recorded neurons of the central clock are interacting with other brain areas.
We have observed that temporal behavioural patterns and the central clock show scale invariant behaviour. With disease and ageing, scale invariance is lost, and also in a brain slice preparation when the clock is not communicating with other brain areas, scale invariance is absent. We conclude that scale invariance emerges at the integrated network level. Understanding how neurons and brain regions communicate, coordinate, synchronize, and collectively respond to signals and perturbations is one of the most intriguing, yet unsolved problems in neuroscience. As the output of the SCN is unambiguously measurable in terms of phase and period, the measurements from the different levels of organization, i.e., the molecular level, the cellular level, the organ level and the behavioural level, can be compared. Current studies are aimed at bridging scales, from the micro to the macro level and vice versa, thereby understanding how properties emerge at each of these levels.
4th seminar (June 26, 2015):
Networks of Interacting Localized Structures and the Process of Desertification
Prof. dr. Arjen Doelman
Although the dynamics of (systems of) partial differential equations (PDE) is intrinsically infinite dimensional, it can remarkably often be reduced to a low-dimensional dynamical system. `Far from equilibrium', such low-dimensional reductions typically govern the positions and interactions between localized structures such as `pulses', `spots' or `stripes'. These systems can be represented as networks, with nodes encoding `particles' (the underlying localized structures) and links encoding the interactions between them. The governing PDE determines the topology of this network, which is fully connected when each `particle' directly interacts with all others. Moreover, there is more going on than `just' the dynamics encoded by this network of `particles and interactions': the network itself may also be dynamical. For instance, `particles' in the network may vanish - in case the respective localized structure becomes unstable due to the PDE dynamics, or may even split into two - when the localized structure bifurcates. This talk - which focusses on the bigger picture while being mostly based on examples - will show how this point of view may be applied for developing a fundamental understanding of the process of desertification, which may be seen as a `massive extinction' of the `particles' associated to localized patches of vegetation (embedded in bare soil).
5th seminar (September 25, 2015):
Ebola. Big Data and Complex Networks in the Real World
Prof. dr. Aske Plaat
In 2014, as the Ebola epidemic in West Africa progressed, the world witnessed in all its horrific glory a complex network in action. Doctors, epidemiologists, complex network scientists, and data scientists have worked hard to counter the disease, and now the outbreak appears to be almost over. From a complex networks perspective the Ebola case affords a possibility to see how our theories hold up, and what there is to be learned. In this talk I will give a brief overview of the Ebola outbreak from a data science perspective, and discuss some of the issues that occurred as science met the real world.
6th seminar (October 30, 2015)
Entropy, inference and economic networks
Dr. J.F. Joao Dias Rodrigues
In this talk I will present E. T. Jaynes' concept of entropy as an inference tool and clarify when and how entropy can be used in the study of complex networks. Entropy is a property which indicates the spread of a probability distribution and has two applications: it can be used to determine a prior distribution when information on constraints is available; and it can be used to infer constraints when the distribution is known. I will review and discuss applications of entropy to the study of the arrow of time, language, food webs and firm sizes, as well as to algorithms of image reconstruction and trade flow estimation and reconciliation. Some of these applications can be interpreted within Jaynes' framework while others cannot. To conclude, I will identify current theoretical and empirical challenges in the study of economic networks and suggest how entropy can help to overcome them. These problems stem from the arbitrary nature of classifications and the aggregation of official statistical trade data, which make the analysis scale-dependent. Entropy can be used to identify correlations among trade flows, trade-offs in the uncertainty of disaggregated data and estimate firm and transaction size distributions.
7th seminar (November 27, 2015)
Bibliometric network analysis: Software tools, techniques, and an analysis of network science at Leiden University
dr. Ludo Waltman and dr. Nees Jan van Eck
We provide an introduction into the research program on bibliometric network analysis at Leiden University's Centre for Science and Technology Studies (CWTS). We demonstrate two popular software tools for bibliometric network analysis developed at CWTS: VOSviewer (www.vosviewer.com) and CitNetExplorer (www.citnetexplorer.nl). We also discuss the techniques that we have developed for network layout and community detection. Finally, we use bibliometric network analysis to study the field of network science and the contributions made to this field by researchers at Leiden University.
Nees Jan van Eck (left) and Ludo Waltman (right)
8th seminar (January 29, 2016)
Mining Large-scale Corporate Networks (slides)
dr. Frank Takes
Corporate networks model the relationships between firms in our economy. Examples of such relationships include ownership ties (firm A owns firms B) or board interlock ties (linking firms based on shared senior level directors). This talk considers the global corporate network, demonstrating results of different studies in which the connectedness of the largest 1 million firms across the globe is investigated. Topics include data quality implications, network topology, centrality analysis and community detection. The results provide interesting insights in the world's most powerful countries and firms, as well as patterns illustrating the network structure of tax havens.
9th seminar (February 26, 2016)
From the brain to the economy: finding communities in networks and correlation matrices
The mesoscopic organization of complex systems, from financial markets to the brain, is an intermediate between the microscopic dynamics of individual units (stocks or neurons, in the mentioned cases), and the macroscopic dynamics of the system as a whole. Indeed, many systems tend to organized in a modular way, with functionally related units being correlated with each other, while at the same time being relatively less (or even negatively) correlated with functionally dissimilar ones. The empirical identification of such emergent organization is challenging due to unavoidable information loss, when inferring the structure from the original time series activity data. In this talk, I will present a modularity based community detection approach for correlation matrices. The method uses maximum-entropy null model designed specifically for correlation matrices (and not networks) that is able to filter out both unit-specific noise and system-wide dependencies. This results in identification of meso-scale functional modules that are internally correlated and mutually anti-correlated. I will present applications to brain networks, financial markets, and international trade. Lastly, using the maximum-entropy framework, I will discuss a new null model for community detection. This "enhanced" null model is able provide link expectations based on both the strengths and the topology of the network. The application of this model to the International Trade Network reveals differences with respect to the standard approach.
10th seminar (April 29, 2016)
Methods & algorithms for detecting communities in large networks
dr. Vincent Traag
Many complex networks have a modular structure: groups of densely connected nodes with few connections between the groups. Nodes in such groups often have something in common, and enrich our understanding of complex networks. Finding such so-called communities in large networks is far from trivial. One of the best-known methods for community detection is modularity, which specifies a quality function of a partition. However, modularity suffers from a well-known flaw, known as the resolution limit: it tends to oversimplify, and lump together several (sub)communities in one large community. We here show that only few quality functions can address this issue. One of the best algorithms for optimising modularity is the Louvain algorithm. We here show that it can lead to arbitrarily badly connected communities---in addition to the resolution limit of modularity. In particular, it can lead to disconnected communities. We here introduce a new algorithm, and show it not only addresses this caveat, but also that it asymptotically ensures that no subset of any community can be moved to another community. Finally, we introduce a fast local move subroutine, speeding up the algorithm 5-10 times.
11th seminar (May 27, 2016)
Ground truth? Clustering scientific publications
dr. Vasyl Palchykov
Community detection techniques are widely used to infer hidden structures within interconnected systems. Despite demonstrating high accuracy on benchmarks, they reproduce the external classification for many real-world systems with a significant level of discrepancy. A widely accepted reason behind such outcome is the unavoidable loss of non-topological information (such as node attributes) encountered when the original complex system is represented as a network.
In this talk we will show that the observed discrepancies may also be caused by a different reason: the external classification itself. For this end we use scientific publication data, which i) exhibit a well defined modular structure and ii) hold an expert-made classification of research articles. Having represented the articles and the extracted scientific concepts both as a bipartite network and as its unipartite projection, we applied modularity optimization to uncover the inner thematic structure. The resulting clusters are shown to partly reflect the author-made classification, although some significant discrepancies are observed. A detailed analysis of these discrepancies shows that they carry essential information about the system, mainly related to the use of similar techniques and methods across different (sub)disciplines, that is otherwise omitted when only the external classification is considered.
In September 2016, the Leiden Networks Day replaced the regular monthly seminar.
12th seminar (October 28, 2016)
Epidemic Spread in Networks (slides)
prof. dr. Piet Van Mieghem (TU Delft)
Epidemic models are increasingly used in real-world networks to understand diffusion phenomena (such as the spread of diseases, emotions, innovations, failures) or the transport of information (such as news, memes in social on-line networks and flows in functional brain networks). We will mainly focus on Susceptible-Infected-Susceptible (SIS) epidemics on networks. After a brief review of the SIS Markovian process on a graph, we will show why SIS epidemics on networks are so interesting and we will overview our recent developments.
13th seminar (November 25, 2016)
Individuals, groups and explorations in networks
dr. Luca Avena