previous day
next day
all days

View: session overviewtalk overview

09:00-09:45 Session Speaker S4: Andrei Zinovyev - Random walks on biological networks
Random walks on biological networks

ABSTRACT. Biological networks represent our knowledge about cellular biochemistry in formalized and simplified form. Random walk with restart on the network graph is currently one of the central methods to explore the biological network structure and to exploit it together with high-dimensional omics data. I will present several applications of this approach to cancer data normalization, classification of samples, defining biological functions. I will describe general formalism to random walks with restart called Google matrix approach, with recent developments for quantification of hidden connections and taking into account edge signs. I will demonstrate, using the example of Wikipedia, that the suggested methods are scalable to graphs with millions of nodes and hundreds of millions of edges.

09:45-10:00 Session Poster P3 A: Economics & Finance
Emergence of Risk Sensitivity from First Principles: An Agent-based model

ABSTRACT. Risk preferences have attracted a large amount of debate in different disciplines. Here, we propose a possible explanation for the origin of risk sensitivity based on the supposition that they can emerge in complex systems from genetic and cultural evolution. An agent-based model was developed to test the plausibility of this hypothesis. In the model, agents survive choosing between a riskier and a safer decision. The contribution of this paper is threefold. First, it provides an overview of risk sensitivity emer-gence in a population of agents. Second, it appraises the effect of environmental factors on this process, and it analyzes the influence of adaptation style on that effect. Finally, it highlights non-monotonic and non-linear re-lationships between the emerging risk preferences and the dangerousness of the in-silico environment. The simplicity of the proposed approach suggests that these findings could apply to different application fields.

Analysis of the Sybil defense of Duniter-based cryptocurrencies

ABSTRACT. Duniter-based cryptocurrencies, which are providing a kind of universal basic income, are using a graph system called Web of Trust to avoid large Sybil attacks. We investigate here the largest size of a Sybil attack depending on the number of attackers and on the parameters of the system.

10:00-11:15 Session Oral O4: Structure & Dynamics
Predictions for dynamics of the World Wide Web

ABSTRACT. Rapid dynamics of the World Wide Web represent a challenge for crawling and indexing web pages. The challenge is encountered on daily basis by focused crawlers in their task to provide businesses with timely and complete information on selected areas of the web. In this work, we introduce prediction models for two metrics that are important in organizing the crawling order of pages: the number of new outlinks, and the change rate of a page. The results show that static page features such as content and text length have high predicting value for change rate and new outlinks. Moreover, the consistency in formation of new outlinks in the provided data results in high quality predictions using only short-history features.

Classical versus Community-aware Centrality Measures: An Empirical Study
PRESENTER: Stephany Rajeh

ABSTRACT. Classical and community-aware centrality measures are two main approaches for identifying influential nodes in complex networks. Nonetheless, both contrast in the way they locate these nodes. This work investigates the relationship between classical and community-aware centrality measures using empirical data. Results demonstrate that the correlation between representative measures of these two approaches ranges from low to medium values. Furthermore, transitivity, efficiency, and mixing parameter are critical network topological properties driving their interactions.

NodeSim: Node Similarity based Network Embedding for Diverse Link Prediction
PRESENTER: Akrati Saxena

ABSTRACT. In complex networks, links can be categorized as intra-community links if both end nodes of the link belong to the same community, otherwise inter-community links. In our work, we propose a network embedding method, called NodeSim, which captures both the similarities between the nodes and the community structure while learning the low-dimensional representation of the network. We verify the efficacy of the proposed embedding method over state-of-the-art methods using diverse link prediction. Extensive experimental results demonstrate the effectiveness of the proposed framework for both inter as well as intra-community link prediction.

Statistical properties of edges and bredges in configuration model networks
PRESENTER: Eytan Katzav

ABSTRACT. A bredge (bridge-edge) in a network is an edge whose deletion would split the network component on which it resides into two separate components. Since the integrity of most networks (particularly transportation and communication networks) is essential for their functionality, bredges are vulnerable links that play an important role in network collapse processes. Therefore, the abundance and properties of bredges affect the resilience of the network to both inadvertent failures and deliberate attacks. It is worth mentioning, however, that there is also a constructive role for bredges, as these are precisely the edges in a network one needs to remove in efficient strategies for containment (islanding) of blackouts in power grids, financial shocks, and vaccination in the context of epidemics. From a different perspective, bredges are fundamental quantities in describing the structure of networks. We present analytical results for the statistical properties of bredges in configuration model networks using based on the cavity method. We examine the distinct properties of bredges on the giant component (GC) and on the finite tree components (FC) of the network. We also consider the fate of bredges in percolation processes.

Applications of renormalizable binary fitness-based network models

ABSTRACT. A recent binary fitness-based renormalizable network models is fit to empirical networks across several levels. First, the case when a "natural" hierarchical partition is available is examined, and examplified with trade data. The goodness of fit at a fixed level is compared to the output of established fitness-based probabilistic models.

Then the goodness of fit of the renormalizable model is assessed when the hierarchical partition is generated by state-of-the art model-based hierarchical clustering methods.

12:15-13:30Lunch Break
13:30-15:15 Session Oral O5: Diffusion/Circulation of Knowledge
The path to scientific discovery: distribution of labor, productivity and innovation incollaborative science
PRESENTER: Maria Castaldo

ABSTRACT. Polymath project constitutes an unprecedented opportunity to understand the path to scientific discovery in an online collaborative environment. To dig into the Polymath experience, in this work, we analyze all the posts related to the projects that arrived to a final peerreviewed publication (projects 1,4,5,8,15) with a particular attention to the organization of labor and the innovation related to author contributions. We observe that an important presence of sporadical contributor boosts the productivity of the most active users and that the production, in terms of number of posts, grows superlinearly with the number of contributors participating to the discussion. When in comes to innovation, we point our that, in large scale collaborations, there is no strict rule determining, a priori, who the main innovators will be. Sometimes, serendipitous interactions by sporadical contributors can have a large impact on the discovery process and a single post by a occasional participant can be responsible of giving a new directionto the work.

Simulating the diffusion of innovation using agent-based models, formal argumentation and the theory of planned behavior

ABSTRACT. Agent-based simulation has long been used to study the dynamics of innovation adoption and diffusion. However, the vast majority of these works are limited to an abstract and simplified representation of this process, which does not make it possible to explain the reasons for an agent's change of opinion, an element that is nonetheless fundamental to understanding the dynamics of innovation diffusion. In order to overcome this limitation, we propose an agent-based model of innovation adoption and diffusion based on a classical theory of psychology, the theory of planned behaviour, and on formal argumentation. Each agent thus has the opportunity to exchange arguments with another agent and to build his/her opinion on an innovation from the set of arguments he/she knows. An application of this model is proposed to study the diffusion of communicating water meters by farmers on the Louts River (South-West of France).

Exploring, browsing and interacting with multi-scale structures of knowledge
PRESENTER: Lobbé Quentin

ABSTRACT. The ICT revolution has given birth to a world of digital traces. A wide number of knowledge-driven domains like science are daily fueled by unlimited flows of textual contents. In order to navigate across these growing constellations of words, interdisciplinary innovations are emerging at the crossroad between social and computational sciences. In particular, complex systems approaches make it now possible to reconstruct multi-level and multi-scale structures of knowledge by means of phylomemies: inheritance networks of elements of knowledge.

In this presentation, we will introduce an endogenous way to visualize the outcomes of the phylomemy reconstruction process by combining both synchronic and diachronic approaches. Our aim is to translate high-dimensional phylomemetic networks into graphical projections and interactive visualizations. To that end, we will use seabed and kinship views to translate the multi-level and multi-scale properties of complex branches of knowledge. We will then define a generic macro-to-micro methodology of exploration implemented within an open source software called Memiescape and validate our approach by browsing through the reconstructed histories of thousands of scientific publications and clinical trials.

The Role of Graphlets as Event Precursors in Social Networks
PRESENTER: Hiba Abou Jamra

ABSTRACT. The increasing availability of data from online social networks (OSN) challenges researchers who seek to build algorithms and machine learning models to analyze users' behaviors and extract knowledge. OSN enhance the emergence of echo chambers where ideas are amplified and can conduct to a digital crisis. The detection of event precursors in these networks can help control and prevent disease outbreaks and global crises.

We present an approach for detecting event precursors in OSN, that aims to generate alerts to help organizations make quick reactions and take preventive decisions against a crisis or an important phenomenon. The method described in this work relies on the analysis of the structure of social networks via graphlets (a particular type of network motifs), to extract and identify event precursors that appear before an important event in these networks. We also study the contribution of these graphlets to the event. After analysis of the experimental results, we show that some graphlets can be considered precursors of events.

Data augmentation impact on domain-specific text summarization
PRESENTER: Abdelghani Laifa

ABSTRACT. This paper aims at presenting a method developed for summarization of French financial texts. Our work highlights the relevance of fine-tuning a pre-trained CamemBERT model using its attention mechanisms on our custom dataset to extend its knowledge on the domain-specific vocabulary, and the impact of textual data augmentation, improving [6%-11%] the performance of our extractive text summarization model. The deep learning methods used, as well as the corpus, are presented here.

SDG achievement patterns: outcome vs expectations in a complex network model

ABSTRACT. We realize a data-driven evaluation of the performance of UN Member States in the achievement of Sustainable Development Goals (SDGs). By constructing complex networks of countries, in which connections are determined by pairwise correlations between vectors of indicators related to a specific SDG, we apply community detection algorithms to partition countries in homogeneous goal achievement groups. We determine the indicators that are more relevant in determining the composition of communities, that can be considered crucial to the achievement of an SDG. We finally compare two networks in which nodes are represented by SDGs, the former based on the similar achievement levels by countries, the latter based on semantic similarities in the statements, to highlight discrepancies between conceptual similarities and performance correlation.

An agent-based model of (food) consumption: Accounting for the Intention-Behaviour-Gap on three dimensions of characteristics with limited knowledge

ABSTRACT. We propose an agent-based model to study purchase diffusion. We conceive goods with 3 dimensions of characteristics, price premiums and limited knowledge on the part of consumers. Despite purchase intention for each characteristic being high, only a reduced number of consumers are initially able to acquire them. The model thus reproduces the Intention-Behaviour-Gap often identified in sustainable food consumption, by exploring two of its known sources: price premiums on the presence of extra characteristics, and lacks in consumers' knowledge as to which goods contain them. We analyse the extent to which knowledge of characteristics diffuses throughout the population and purchases of them are adopted. By testing how different parameters (knowledge spillovers, average network degree, knowledge availability and price premiums) affect these evolutions, we offer insights as to how wider adoption of desirable purchase behaviours can be encouraged. Results show that all parameters have significant effects on knowledge availability and purchase behaviour, and that the ensuing increased knowledge particularly affects purchases of 2 and 3 characteristics. Modifying network parameters (average network degree and knowledge spillovers) produces effects comparable to those of external ones (initial knowledge availability and price premiums), an interesting feature in terms of policy recommendations since the former can arguably imply less costly interventions than the latter.

15:15-15:30 Session Poster P4 A: Social Complexity
Populations preferences through Wikipedia edits: A dynamic analysis
PRESENTER: Julien Assuied


Assessing movie similarity using a multilayer network model
PRESENTER: Majda Lafhel

ABSTRACT. Recommender systems have been used in many different applications, especially in movie recommendation which is a productive domain for recommendation technology. Recently, most of data sets is produced from movie domain such as Netflix competition, and they have suggested many works on movie recommendation. However, the majority of the existing approaches use rating information to generate recommendations. What is missing from these approaches is they not consider the movie story in the recommendation process which can encompass a wide range of information of movie content, such as the characters, time and locations which can be used to compare movies. In the past years, network models have been increasingly used for supporting analysis of stories, such as books, famous TV series, news, and movies. Most of these network models focus on one facet of the movie story, most of time characters, and capture all their interactions, to produce a global picture of the story’s content. Other works went beyond by introducing other semantic elements such as dialogues. However, they always captured the information in a single layer network or a bipartite graph. To deal with that, we proposed in our previous work \cite{mourchid} a multilayer network model that capture more elements of the movie story compared to regular networks. It encompasses the single network analysis based either on characters or scenes and proposes new topological analysis tools. In order to compare networks using semantical elements such as characters, locations, and keywords, we propose to exploit this multilayer network as a model for computing similarities. A difficult problem when studying networks is that of comparison. To deal with that we propose to use the proposed method by Bagrow \emph{et al.}\cite{bagrow} to characterize network similarity. For each network, the authors compute its B-matrix which is considered a signature that represents the network and serves as its 'portrait'. Results demonstrate that computing the similarity between networks using their portraits corroborates human similarity perception.

15:30-16:15 Session Speaker S5: Cesar A. Hidalgo - How Humans Judge Machines
How Humans Judge Machines

ABSTRACT. How would you feel about losing your job to a machine? How about a tsunami alert system that fails? Would you react differently to acts of discrimination performed by a machine or a human? How about public surveillance? How Humans Judge Machines compares people’s reactions to actions performed by humans and machines. Using data collected in dozens of experiments, this book reveals the biases that permeate human machine interactions. Are there conditions in which we judge machines unfairly? Is our judgment of machines affected by the moral dimensions of a scenario? Is our judgment of machine correlated with demographic factors, such as education or gender? Hidalgo and colleagues use hard science to take on these pressing technological questions. Using randomized experiments, they create revealing counterfactuals and build statistical models to explain how people judge A.I., and whether we do it fairly or not. Through original research, they bring us one step closer to understanding the ethical consequences of artificial intelligence. How Humans Judge Machines can be read for free at (in print with MIT Press).

16:30-17:45 Session Oral O6: Biological & Ecological Networks
A Complex Network Analysis to modeling Ecological Networks
PRESENTER: Enrico Podda

ABSTRACT. There has always been a certain level of complexity in the study of the territory. For as long as we can remember, humankind has shaped the environment around it to satisfy its needs. This has led to great achievements, but not for many animal or plant species. Modern infrastructure has destroyed environments upon which plants and animals depended for survival. Nations have become aware of this in recent years, and have begun to mobilize for more responsible management of the territory and the species that inhabit it. However, although the areas used for natural habitats are increasingly used for the protection of species, protected areas alone cannot assist species in every aspect of their survival, especially regarding terrestrial animal species. This paper presents an analysis of the study of the territory through complex networks. The analysis will focus on algorithms designed to predict and facilitate the migration of terrestrial species between natural areas, some of which are already known, while others are being implemented.

Predicting invasion by polarized migrating cells

ABSTRACT. In many circumstances, including development, wound healing, and cancer growth, cells migrate and tend to invade space. In the case of cells from diffuse glioma, this migration occurs on very large distances (several centimetres) away from the tumour and leads to the existence, in the brain of the patient, of slightly infiltrated regions. They can't be detected by routine imaging techniques because the concentration of tumour cells is too low, and they are the cause, in the majority of cases, of recurrences after treaments which ultimately bring about a fatal issue. Therefore, an accurate quantitative prediction of invasion patterns of tumour cells by a mathematical model would be helpful to guide treatments and enhance the prognosis.

Simple models, belonging to the family of agent-based models or cellular automata, are appealing for this task, and many have been suggested for populations of migrating cells. This approach can be completed by the derivation of continuum models (usually one or a few coupled partial differential equations) which are better suited for predictions at the organ scale than direct simulations of sets of several hundred millions of cells.

However, models proposed so far did not take into account some fundamental and ubiquitous feature of the migrating eukaryotic cells, both in 2D and in 3D environments, namely the mechanism of polarisation (when migration starts) and depolarisation (when migration stops or when the cell changes direction)~\cite{Valencia15}. During these transformations, the morphology of the cell changes, which can be the cause of steric constrains.

Here we explore the consequences on collective cell migration of these phenomena, first at the scale of a few cells or a small tissue thanks to simulations of a cellular automaton, then at the scale of the tumour or the organ by taking the continuum limit of our model. We derive a simple (yet approximate) nonlinear diffusion equation for the concentration of tumour cells, the solutions of which are good approximations of the predictions of the cellular automaton.

An agent-based model to simulate co-infection with human papillomaviruses in a partially vaccinated partnership network

ABSTRACT. Human papillomaviruses (HPV) are among the most common sexually transmitted infec-tion and a necessary cause of cervical cancer. In the context of vaccination against a sub-group of genotypes, better understanding the role of biological interactions between HPV genotypes and social interactions between humans is essential to anticipate what the vaccine impact could be at the population level. Existing models that study interactions between genotypes are based on basic homogeneous assumptions about human-to-human transmission, whereas agent-based models that take into account the heterogene-ity of human behavior and transmission do not consider possible interactions between genotypes. Here, we present a novel stochastic agent-based model formalizing the co-circulation on a human partnership network of multiple interacting genotypes, some of them being preventable by the vaccine (vaccine types) and others not. The model explicit-ly formalizes heterogeneity in sexual behaviors and makes it possible to explore distinct genotypic interaction mechanisms during intra-host co-infections. Using model simula-tions, we investigate infection dynamics in the population after vaccine introduction de-pending on assumptions regarding vaccine coverage and vaccine and non-vaccine geno-type interactions.

Statistical properties of the MetaCore network of protein-protein interactions

ABSTRACT. The MetaCore commercial database describes interactions of proteins and other chemical molecules and clusters in the form of directed network between these elements, viewed as nodes. The number of nodes goes beyond 40 thousands with almost 300 thousands links between them. The links have essentially bi-functional nature describing either activation or inhibition actions between proteins. We present here the analysis of statistical properties of this complex network applying the methods of the Google matrix, PageRank and CheiRank algorithms broadly used in the frame of the World Wide Web, Wikipedia, the world trade and other directed networks. We specifically describe the Ising PageRank approach which allows to treat the bi-functional type of protein-protein interactions. We also show that the developed reduced Google matrix algorithm allows to obtain an effective network of interactions inside a specific group of selected proteins. This method takes into account not only direct protein-protein interactions but also recover their indirect nontrivial couplings appearing due to summation over all the pathways passing via the global bi-functional network. The developed analysis allows to espablish an average action of each protein being more oriented to activation or inhibition. We argue that the described Google matrix analysis represents an efficient tool for investigation of influence of specific groups of proteins related to specific diseases.

Respiratory influences on heart rate variability self- organized criticality

ABSTRACT. Self-organized criticality (SOC) of the cardiovascular function has recently been demonstrated in the standing human. Some sequences of heart beats called bradycardia sequences are unexpected because of the strong cardiovascular regulation. Bradycardia sequences are distributed across a straight line in log-log graph according to Zipf’s law, a feature of SOC. Unexpectedly, Zipf’s law is broken at a delay of five beats. This delay matches the delay of spontaneous respiratory influences on the cardiovascular system. We hypothesized that cardiovascular SOC is influenced by breathing. To check this hypothesis, we recorded beat-by-beat heart rate of six standing healthy subjects with imposed breathing patterns: spontaneous pattern as a control, breathing frequency of 0.1, 0.2, and 0.25 Hz. We counted and classified bradycardia episodes according to their length in number of beats. We checked whether these episodes are distributed according to Zipf’s law. Best linear fit identified the broken law tipping point. Bradycardia sequences were distributed according to straight lines in all the subjects but one. The tipping points were at a delay of four beats. Imposed breathing pattern have no effect on the tipping point. We do not explain the difference of the tipping point position with previous studies. Breathing does not influence cardiovascular SOC. Alternative explanations include influences of arterial compliance or of sympathetic nervous system delay. Another explanation could be a finite size effect.