Seminars are usually held on Friday from 11:00 to 12:00 in the main seminar room (Euler Room - A002) of the Building Euler.
If you wish to receive the seminar announcements, please send an email to
The subject line should be 'subscribe big-data yourname' where yourname is your first and last name.

Thursday, June 11, 11:00
Turning Big Data into Social Impact: Challenges and Opportunities
Robert Kirkpatrick, Director of UN Global Pulse

The emergence of big data has already had a profound impact on the business world, transforming business models across sectors and spawning a new technological arms race around real-time extraction of insights. At the same time, the public is increasingly aware of the fact that governments are mining ever more data in the name of national security, big leading many to conclude that the mere existence of big data creates unacceptable risks to individual privacy.

In this seminar, UN Global Pulse director Robert Kirkpatrick will elaborate the potential of big data to serve the public good, strategies to mitigate the risks of harms, and the efforts of the United Nations to discover, develop and evaluate applications of data science to address priority challenges in sustainable development, climate change resilience and humanitarian action.

Tuesday, June 9, 12:00
Stemming the global Tuberculosis and MDR-TB epidemic
Aamir Khan, Interactive Research & Development (IRD), Pakistan and South Africa

Dr Aamir Khan is a medical doctor and epidemiologist based in Karachi, Pakistan. He is the Executive Director of Interactive Research & Development (IRD), a research and service delivery organisation committed to improving the health of vulnerable communities.

Tuberculosis, or TB, is an infectious bacterial disease caused by Mycobacterium tuberculosis. It is transmitted via droplets from lungs of people with the active disease. Each year, 9 million people get infected. 1.3 million die as a consequence of the limited access to health services and treatment. Due to sub-optimal TB control in fragile health systems, the emergence of drug-resistant TB (MDR-TB) has become a global threat which particularly affects eastern-european and asian countries.

With the support of the World Health Organization (WHO) and the Stop TB partnership, he coordinates the implementation of mass screening strategies for Tuberculosis (TB) in Asian megacities and in the South-Africa mining sector. With his team, Aamir applies the most recent social, medical and technological innovations within these screening programs. His team has developed e-Health tools which allows to build unprecedented database collected from the mass strategies. This information, and the new knowledge that emerges from it, are critical in the global effort to stem the TB and MDR-TB epidemics.

Dowdy, D. W., et al. (2013). "Population-level impact of active tuberculosis case finding in an Asian megacity." PLoS One 8(10): e77517.
Fraser, H. S., et al. (2013). "E-health systems for management of MDR-TB in resource-poor environments: a decade of experience and recommendations for future work." Stud Health Technol Inform 192: 627-631.
Khan, A. J., et al. (2012). "Engaging the private sector to increase tuberculosis case detection: an impact evaluation study." Lancet Infect Dis 12(8): 608-616.

Tuesday, June 9, 11:00
Avalanche-Outbreaks Emerging in Cooperative Contagions
Fakhteh Ghanbarnejad, Robert Koch Institute, Berlin

Spreading of contagious agents can exhibit a percolation transition that separates transitory prevalence from outbreaks which reach a fi nite fraction of the population [1, 2]. Such transitions are commonly believed to be continuous, featuring a mild onset of outbreak at the threshold. Empirical studies have however shown drastically more violent spreading modes when the participating agents are not limited to one type. The striking examples are the co-epidemic of the Spanish flu and pneumonia that occurred in 1918 [3, 4], and more recently, the concurrent prevalence of HIV/AIDS and a host of diseases [5-7]. Yet it is still not clear to what extent such interaction of pathogens could alter the scenario of outbreak from an ordinary single-agent process. Here we study a mechanistic model for understanding contagion processes involving inter-agent cooperations. Our stochastic simulations reveal the possible emergence of a massive avalanche-like outbreak right at the threshold, which is manifested as a discontinuous phase transition. Such an abrupt change arises only if the underlying network topology supports a bottleneck for cascaded mutual infections. Surprisingly, all these discontinuous transitions are accompanied by nontrivial critical behaviours, presenting a rare case of hybrid transition [8]. The findings may imply the origin of catastrophic occurrences in many realistic systems, from co-epidemics to financial contagions [9].

[1] Stauffer, D. & Aharony, A. Introduction to Percolation Theory (Taylor & Francis, London, 1994).
[2] Newman, M. Networks: An Introduction (Oxford University Press, New York, 2010).
[3] Brundage, J. F. & Shanks, G. D. Deaths from bacterial pneumonia during 1918-19 influenza pandemic. Emerg. Infect. Dis. 14, 1193-1199 (2008).
[4] Taubenberger, J., & Morens, D. 1918 Influenza: the mother of all pandemics. Emerg. Infect. Dis. 12, 15-22,(2006).
[5] Pawlowski, A., Jansson, M. Skold, M., Rottenberg, M. E. & Kallenius, G. Tuberculosis and HIV co-infection. PLOS Pathogens 8, e1002464 (2012).
[6] Chang, C. C., et al. HIV and co-infections. Immun. Rev. 254, 114-142 (2013).
[7] Petney, T. & Andrews, R. Multiparasite communities in animals and humans: frequency, structure and pathogenic significance. International Journal for Para-sitology 28 377 (1998).
[8] Goltsev, A. V., Dorogovtsev, S. N. & Mendes, J. K-core (bootstrap) percolation on complex networks: Critical phenomena and nonlocal effects. Phys. Rev. E 73, 056101 (2006).
[9] Claessens, S, & Forbes, K. (Eds.) International Financial Contagion: An Overview of the Issues (Springer, NewYork, 2009)

Thursday, June 4, 11:00
A Riemannian approach to large-scale constrained least-squares with symmetries
Bamdev Mishra, University of Liège

The talk deals with least-squares optimization on a manifold of equivalence relations, e.g., in the presence of symmetries which arise frequently in many applications. While least-squares cost functions remain a popular way to model large-scale problems, the additional symmetry constraint should be interpreted as a way to make the modeling robust. We discuss two fundamental, and recently popular, examples of the matrix completion problem, a least-squares problem with rank constraints and the generalized eigenvalue problem, a least-squares problem with orthogonality constraints.

The possible large-scale nature of these problems demands to exploit the problem structure as much as possible in order to design numerically efficient algorithms. To this end, we propose the notion of Riemannian preconditioning with metric tuning.

For the generalized eigenvalue problem, our approach connects to the classical power, inverse, and the Rayleigh quotient iterations, all interpreted as Riemannian steepest descent algorithms with specific metric choices. For the matrix completion problem, our approach leads to metric choices that connect and compete effectively with state-of-the-art.

Tuesday, May 12, 11:00
Genesis of millet prices in Senegal: the role of production, markets and their failures
Damien Jacques, UCL

Stable prices are the main indicator of food access and a key determinant of the revenues of those living in agricultural zones. Differentials in prices between producing (low prices) and consuming (high prices) areas harm both groups and indicate the presence of market failures.

In this study we model the millet prices formation process in Senegal in a spatially explicit model that accounts for both high transportation costs and information asymmetries. The model integrates a unique and diversified set of data in a framework that is coherent with the economic theory. The high ability of the model in reproducing the price differentials between 41 markets (r 2 > 80%) opens a new avenue for the research on market integration which (i) integrates production data derived from remote sensing, (ii) simulates the demand and supply at the local level and (iii) the arbitrage process between imperfectly integrated markets.

Friday, May 8, 14:00
Functional Similarity Analysis and Deep Learning for Multimedia Indexing
Leonardo Gutierrez, Université Joseph Fourier, Grenoble

This talk is divided in two sub-talks.

Functional Similarity Analysis:

Systematic performance improvement across hardware and software driven resources remains a veritable challenge in increasingly complex system-on-chip development projects. Meanwhile, graphical networks analysis based on mathematical graph theory offers a heuristic approach to dynamic profiling at source code level and instruction code level. To improve the visually interactive heuristic approach, a rule set of functional similarity between code segments represented by hierarchical graphs was elaborated, and compared with the effectiveness of an already existing customizable processor architecture platform. In addition, a histogram of functional similarity was established upon graphical networks composed of several different software programs.

Deep Learning for Multimedia Indexing:

Automatic indexing of image and video documents is a difficult problem because of the "distance" between arrays of numbers encoding their content and concepts (e.g. people, places, events or objects) with which users want to annotate. State of the art methods generally rely on an extraction / classification / fusion pipeline. The content of documents is first extracted in the form of descriptors of fixed size. Recently, "deep learning" or "deep convolution networks" methods have proven to be a very effective alternative. These methods include both the feature extraction and classification steps in a single and consistent approach. These networks learn all relevant descriptors both without requiring fusion. They also take into account the implicit relationships between concepts. Although they may achieve similar and generally better than conventional approaches performance, some studies have shown that they are complementary with them, and they can be combined with them for producing systems with even greater performance.

Talk in French with slides in English

Tuesday, March 31, 14:00
Bayesian networks, Information-theoretic networks and Probabilistic logic
Patrick Meyer, ULg

In this talk, causality concepts that are used in Bayesian networks inference will be connected to information-theoretic network inference algorithms used in Systems Biology. The latter approaches have a good trade-off accuracy-computational complexity. However, a promising improvement, centered around a statistical version of the logical implication, will be introduced in the last part of this talk.

Thursday, March 5, 11:00
A review of some recent research topics in numerical linear algebra
Vanni Noferini, University of Manchester.

I will first give an overview of some research themes I have been interested in during recent years, including matrix polynomials, numerical root-finding, matrix functions, and structured linear algebra.

In the second part of the talk I would like to present more in detail an efficient resultant-based bivariate root-finder [1] that has been implemented in the open-source software package Chebfun [2].

[1] Y. Nakatuskasa, V. Noferini, A. Townsend. Computing the common zeros of two bivariate functions via Bézout resultants, Numerische Mathematik, Vol. 129 (1), pp.181-209, 2015.
[2] L. N. Trefethen et al. Chebfun version 5.1. Software, 2014. The Chebfun Development Team.

Tuesday, December 9, 14:00
The importance of non-Markovianity in temporal networks
Antonios Garas, ETH Zurich.

Typically, real complex systems have topologies that evolve with time. This poses limitations to the usual way of studying them through static, time-aggregated networks. In this talk I will discuss about a recently introduced temporal-topological feature called betweenness preference, which describes the tendency of nodes to preferentially connect - in a temporal sense - particular pairs of neighbors. This preferential connectivity results to specific ordering of interactions in temporal networks, alters causality, makes the system non-Markovian, and affects the evolution of dynamical processes. Thus, I will describe a methodology to analytically predict causality-driven changes of diffusion speed in non-Markovian temporal networks. I will show that compared to the time-aggregated network, non-Markovian characteristics can lead to both a slow-down or speed-up of diffusion, which can even outweigh the decelerating effect of community structures in the static topology. Therefore, I will argue that non-Markovian properties of temporal networks constitute an important additional dimension of complexity in time-varying complex systems.

Tuesday, December 2, 14:00
Geometric methods for recommender systems
Pierre-Antoine Absil, UCL.

The central topic of this talk is low-rank optimization, where the archetypal problem consists of minimizing a real-valued function defined on a set of matrices of fixed or bounded rank. The fact that the set of fixed-rank matrices admits Riemannian manifold structures endows the problem with a rich geometry. We will see how geometric concepts can be exploited to design efficient low-rank optimization methods. Upstream, this will lead us into the realm of Riemannian optimization. Downstream, we will glance through various applications of low-rank optimization, such as the maximal cut of a graph, sparse principal component analysis, and recommender systems, with an emphasis on the latter. In recommender systems, a collection of items are available to users. For example, as popularized by the Netflix prize, the items can be movies and the users can be customers. Each customer has rated some movies, and the task is to predict how much the customers would like the movies they did not rate, so as to make a personalized recommendation. One popular model posits that the matrix of ratings is approximately low-rank. This results in the mathematical problem of finding a matrix of low rank that is optimal, in the sense that it agrees as well as possible (according to some criterion) with measured entries. We will see that, while the best known methods for recommender systems take the form of meta-algorithms that aggregate results provided by various techniques, low-rank optimization for recommender systems has recently made important progress and can provide a useful basis for better predictors. This talk is based on joint work with Nicolas Boumal.

Thursday, November 6, 11:00
Community structure in temporal multilayer networks
Marya Bazzi, Oxford University.

An important feature in many networks is the existence of "communities", sets of nodes that are 'more densely' connected to each other than to nodes in the rest of a network. We investigate a generalization of "modularity maximization", a global community detection method, to temporal networks represented as "multilayer networks". As a focal example, we study time-dependent financial-asset correlation networks.

To represent the temporal dimension of a network in a multilayer framework, one can adopt a choice of inter-layer connection that is uniform and "ordinal". We investigate this choice of representation in a generalization of modularity maximization to multilayer networks. "Modularity" is a function that measures the quality of a partition of nodes in a network by comparing edge weights within sets in the observed network to edge weights within sets in a "null network", generated from a specified "null model". We introduce a diagnostic to measure "persistence" in a multilayer partition, an indication of change in community structure through time. We prove some results to show how multilayer modularity maximization reflects the trade-off between static community structure within layers and higher values of persistence across layers. We discuss an issue that the "Louvain algorithm" faces when applied to multilayer networks with uniform and ordinal inter-layer connections, and we suggest a way to mitigate it. Our results extend to maximization problems where the quality function is not the modularity quality function, provided the resulting maximization problem has the same form.

Thursday, November 6, 11:30
Think Locally, Act Locally: The Detection of Small, Medium-Sized, and Large Communities in Large Networks
Lucas Jeub, Oxford University.

It is common in the study of networks to investigate intermediate-sized (or "meso-scale") features to try to gain an understanding of network structure and function. For example, numerous algorithms have been developed to try to identify "communities", which are typically construed as sets of nodes with denser connections internally than with the remainder of a network.

We adopt a complementary perspective that "communities" are associated with bottlenecks of locally-biased dynamical processes that begin at seed sets of nodes, and we employ several different community-identification procedures (using diffusion-based and geodesic-based dynamics) to investigate community quality as a function of community size.

Using empirical and synthetic networks, we identify several distinct scenarios for "size-resolved community structure" that can arise in real (and realistic) networks:
(i) the best small groups of nodes can be better than the best large groups (for a given formulation of the idea of a good community);
(ii) the best small groups can have a quality that is comparable to the best medium-sized and large groups; and
(iii) the best small groups of nodes can be worse than the best large groups. Which of these three cases holds for a given network can make an enormous difference when investigating and making claims about network community structure, and it is important to take this into account to obtain reliable downstream conclusions. Depending on which scenario holds, one may or may not be able to successfully identify "good" communities in a given network (and good communities might not even exist for a given community quality measure), the manner in which different small communities fit together to form meso-scale network structures can be very different, and processes such as viral propagation and information diffusion can exhibit very different dynamics.

Friday, October 24, 11:00
Estimating Food Consumption and Poverty Indices with Mobile Phone Data
Adeline Decuyper, UCL.

Recent studies have shown the value of mobile phone data to tackle problems related to economic development and humanitarian action. In this research, we assess the suitability of indicators derived from mobile phone data as a proxy for food security indicators, and poverty indices.

Our results show surprisingly good agreement between mobile phone indicators and food security, opening several questions for further research on the subject.

Friday, October 24, 11:30
Burstiness, efficiency and transmissibility of infectious contacts in temporal networks
Martin Gueuning, Université de Namur.

In this work, we are looking at a diffusion process on a network of agents where the probability of success depends on the allocated time for the attempt. We consider different time-allocation strategies for the agents consisting in a trade-off between many unlikely attempts versus few likely ones. Our model incorporates a bursty behaviour as observed in human-related networks such as Twitter or mailbox checking or face-to-face contact patterns. Our results show that, for the same mean time between two successful transmissions, the former strategy is more efficient in terms of diffusion. This applies to marketing strategies as well as to the study of infectious diseases.

Friday, October 10, 11:00
Matrix factorization techniques applied to fMRI-EEG coupled datasets
Matthieu Genicot, UCL.

Useful links : UCLouvain | ICTEAM | INMA