
On the 19th May, 2025, Alejandro de la Concha has successfully defended his PhD in Applied Mathematics at the ENS Paris Saclay. The title is “Graph-based machine learning for detection tasks on complex systems” and I had the pleasure to co-supervise him with Nicolas Vayatis, by working very closely with him for 5 years!
Despite the ups and downs of any such long and transformative process for young researchers, the outcome should make proud Alejandro for its high quality, technical level, methodological and conceptual contribution, and publication production.
Big thanks also to the reviewers of the thesis manuscript and the jury members of the defense:
- Cedric RICHARD (Professor, Université Côte d’Azur)- reviewer
- Taiji SUZUKI (Professor, University of Tokyo) – reviewer
- Agnès, DESOLNEUX (Directror of research, ENS Paris- Saclay)
- Céline LÉVY-LEDUC (Proffesor, Université Paris Cité)
- Gilles BLANCHARD (Professor, Université Paris-Saclay)
“Computer Science is the future”, they used to say.
And despite being still emphatically present today, and will be also in the foreseeable future, at the same time we cannot but notice that things change, especially for that CS part closer to Computing. In some sense, this is for all the range of information sciences.*
The more things get industrialized and automatized, there are two paths for one to choose from. The first is to invest in making this happen, meaning to ride the technological wave, be among the first who does this or that, intensify production, etc., and get your share from the reduction of production costs.
The alternative is to give more emphasis to intellectual depth and width, to promote creativity, the making of fundamental ideas, and the foundation for many of those -in the scientific world- is the cultivation of our mathematical culture.
Food for thought as this affects us in all levels, institutional and personal.
—
- One of the important factors for the drop presented in the plot is that coding (programming, familiarity and accesibility to dev infrastructures) has gradually becoming part of the typical skillset for a range of intersecting fields, mathematicians, physicists, engineers… I remember when, not more than 10 years ago, we had to introduce our maths students to basic computing tools (e.g. how to use an OS, how to write Latex, how to create slides), and now most of them know those things already when entering the school… Of course those are not professional computer scientists or software developers, but the spread of the coding culture means that many coding tasks can be handled by experts of the field, and in a more comprehensive way regarding the ends of a project (rather than taking a coder and try to explain him the purpose of what he is asked to do).
vendredi 4 avril 2025, 9h45 – 18h, salle A4-32 (4è étage), EHESS 54, bd Raspail, 75006 Paris
séance spéciale du “Séminaire Systèmes complexes en sciences sociales“
Journée co-organisée par le CAMS (EHESS-CNRS) et le Centre Borelli (ENS Paris-Saclay-CNRS)
avec le soutien de l’ANR ReaCh.
Programme provisoire
- 9h45 accueil
- 10h – 11h : exposés courts (20mn + 10mn de questions/discussion)
- 10h-10h30 Ferdinand Le Coz
- 10h30-11h Amina Azaiez
- 11h-11h15 pause
- 11h15-12h30 Fabien Tarissan (ENS Paris-Saclay) – exposé de 50mn + 10 à 20mn de questions/discussion
- pause déjeuner
- 14h30-15h45 Paul Guille-Escuret (Institut Pasteur)- exposé de 50mn + 10 à 20mn de questions/discussion
- 15h45-16h pause
- 16h-17h : exposés courts (20mn + 10mn de questions/discussion)
- 16h-16h30 Chiara Giaquinta
- 16h30-17h Gaspard Abel
Lieu : EHESS, 54 bd Raspail, Salle A4-32 (4è étage)
Inscription : participation sur place ou en ligne, veuillez vous inscrire ici. A la demande “Choisissez votre séminaire”, saisir UE499 – attention, pas d’espace entre UE et 499 -, et sélectionner le séminaire. Une réponse automatique vous demandera de confirmer votre demande. La précision “participation obligatoire en présentiel” concerne uniquement les étudiants inscrits à une formation de l’EHESS.
Comité d’organisation :
Gaspard Abel (Centre Borelli & CAMS), Henri Berestycki (EHESS, CAMS), Argyris Kalogeratos (ENS Paris-Saclay, Centre Borelli), Jean-Pierre Nadal (CNRS, LPENS & CAMS), Julien Randon-Furling (ENS Paris-Saclay, Centre Borelli), Camille Roth (CNRS & EHESS, CAMS), Annick Vignes (INRAE & ENPC, CAMS & LISIS).
Contact : jpnadal@ehess.fr
Titres et résumés
(English versions at the bottom of page)
Exposés longs
- Fabien Tarissan (ENS Paris-Saclay)
Titre : Recommandation algorithmique et diversité de l’information. Comment analyser l’impact des algorithmes en ligne ?
Résumé : Que ce soit des problématiques de classement de l’information (moteurs de recherche) ou celles de la recommandation de contenus (via les réseaux sociaux notamment), les algorithmes sont au cœur des processus de sélection qui rendent visibles certaines informations en ligne. Ces choix algorithmiques ont de facto un fort impact sur l’activité des utilisateurs et, dès lors, sur leur accès à l’information. Ceci soulève la question de notre capacité à mesurer la qualité de ces choix algorithmes ainsi que leur impact sur les utilisateurs.
Dans cet exposé, je montrerai comment les structures relationnelles décrivant l’activité d’utilisateurs en ligne peut être analysées et modélisées afin de révéler la diversité de l’information à laquelle ils sont exposés.
- Paul Guille-Escuret (Institut Pasteur)
Titre : Faire ses « propres recherches » pour lutter contre la désinformation, sociologie d’un vigilantisme numérique en faveur de la vaccination.
Résumé : Les controverses vaccinales restent largement représentées comme un combat opposant les autorités sanitaires à des militants anti-vaccins profanes devenus un symbole des rapports problématiques que les populations entretiendraient aux technologies et à la science. Bien avant la pandémie de Covid-19, les acteurs critiques de la vaccination étaient pourtant déjà aux prises avec un contre-mouvement présenté comme une réponse citoyenne face au manque de régulation de la désinformation en ligne. À partir d’une ethnographie menée au sein de collectifs Facebook et d’entretiens réalisés avec leurs modérateurs, une part importante de mon travail de thèse donne à voir les coulisses de ce militantisme numérique « pro-vaccins », ses rétributions et ses tensions normatives. Si mes résultats montrent que ces mobilisations ne sont pas réductibles à leurs composantes antagonistiques, ils attirent aussi l’attention sur l’émergence d’une défense de la Science de plus en plus déconnectée des institutions scientifiques. Tout comme ceux qu’ils désignent comme adversaires, ces militants peuvent être des autodidactes, des profanes qui « font leurs propres recherches ».
Exposés courts.
- Amina Azaiez (Univ. Paris 1 Panthéon-Sorbonne, Centre d’Économie de la Sorbonne) – doctorante avec Antoine Mandel.
Titre : Analyse du réseau de lobbying de la Commission européenne par une modélisation en hypergraphe.
Résumé : Notre étude explore dans quelle mesure le réseau de lobbying auprès de la Commission européenne (CE) est dominé par un cercle restreint d’acteurs bien connectés et comment l’influence se distribue entre les différents types d’acteurs. Nous utilisons un ensemble de données exhaustif et accessible au public, comprenant plus de 18 000 réunions organisées dans le cadre d’un processus de consultation des parties prenantes. Nous modélisons ces interactions en face à face en construisant un hypergraphe dont les nœuds représentent les représentants de la CE et les parties prenantes, tandis que les hyperliens relient les entités participant aux mêmes réunions. Notre analyse met en évidence une structure hiérarchique de type centre-périphérie, où un nombre restreint d’entités bien connectées occupent le centre du réseau et bénéficient d’une intégration stable dans le processus décisionnel de la CE. L’analyse de la composition du centre révèle que les entreprises et les associations professionnelles entretiennent des relations plus proches avec la CE. Une analyse de régression montre que les efforts de lobbying et la taille des entreprises sont des prédicteurs significatifs de la centralité des entreprises, indépendamment d’autres caractéristiques objectives. Nos résultats fournissent des preuves quantitatives soutenant l’idée que le lobbying est un outil dominé par des acteurs bien connectés, tout en mettant en lumière des stratégies de lobbying variées selon les catégories d’acteurs.
- Ferdinand Le Coz (IRIT, Toulouse)
Titre : Unsupervised detection of persistent communities in dynamic networks using Networks embeddings.
Résumé : Je m’intéresse aux Réseaux de Neurones de Graphes (GNN) pour l’étude de réseaux dynamiques. Je propose algorithme pour détecter les communautés dynamiques à partir d’embeddings de graphes sans connaissance à priori de la structure de ces communautés. J’exposerai brièvement une méthode de génération de graphes dynamiques et de features pour tester l’algorithme et discuter de l’impact des features sur les représentations produites par les GNNs. Enfin, je présenterai un cas d’étude où j’essaie de retrouver les communautés politiques sur Twitter à partir des Retweets entre utilisateurs lors d’une campagne présidentielle en France.
- Gaspard Abel (ENS Paris-Saclay, Centre Borelli & CAMS) – doctorant avec Argyris Kalogeratos, Julien Randon-Furling & Jean-Pierre Nadal.
Titre : Analyse de l’activité des réseaux sociaux à l’aide de processus temporels ponctuels conjoints pour l’interaction entre l’utilisateur et le sujet.
Résumé : Les media sociaux ont radicalement modifié la manière dont les gens perçoivent les flux d’informations auxquels ils sont exposés, et la propagation d’une multitude d’informations ou de comportements dans une population est souvent interconnectée. Dans ce travail, nous introduisons le modèle MIC (Mixture of Interacting Cascades), qui est basé sur des processus de Hawkes multidimensionnels marqués. Le modèle MIC décrit conjointement les interactions non triviales entre les utilisateurs et les cascades d’informations. Nous introduisons notamment une visualisation en couche afin de révéler le couplage entre la structure du réseau social et les alignements de sujets.
Titles and abstracts – English versions
- Fabien Tarissan (ENS Paris-Saclay)
Title: Recommendation algorithms and information diversity: how to analyze its impact in online platforms ?
Abstract: Whether it be in the context of information ranking (e.g. search engines) or content recommendation (on social networks for instance), algorithms are at the core of processes selecting which information is made visible. Those algorithmic choices have de facto a strong impact on user’s activity and therefore on their access to information. This raises the question of measuring the quality of the choices made by algorithms and their impact on the users.
In this presentation, we will discuss how to exploit the network structure generated by user’s activity in order to reveal the diversity of the information they access.
- Paul Guille-Escuret (Institut Pasteur)
Title: Doing their “own research” to fight misinformation, sociology of a digital vigilantism in favor of vaccination.
Abstract: Vaccine controversies are largely portrayed as a struggle between health authorities and anti-vaccine activists, who have become a symbol of the problematic relationship that people maintain with technology and science. However, long before the Covid-19 pandemic, critics of vaccination were already facing a counter-movement presented as a civic response to the lack of regulation regarding online misinformation. Drawing from ethnographic research conducted within Facebook groups and interviews with their moderators, a key part of my doctoral work takes a behind-the-scenes look at this digital “pro-vaccine” activism, its rewards, and its normative tensions. While my findings indicate that these mobilizations cannot be reduced to their antagonistic side, they also draw attention to the emergence of a defense of Science that is increasingly disconnected from scientific institutions. Just like those they identify as adversaries, these activists can be self-taught individuals, laypersons who “conduct their own research”.
Short talks
- Amina Azaiez (Univ. Paris 1 Panthéon-Sorbonne, Centre d’Économie de la Sorbonne)
Title: A Hypergraph Analysis of the European Commission Lobby Network
Abstract: Our work addresses the research question of whether the European Commission (EC) lobby network is dominated by a select group of well-connected actors and how influence is distributed among different types of actors. We use a comprehensive, publicly available dataset encompassing over 18,000 meetings as part of a stakeholder consultation process. We model these face-to-face interactions by constructing a hypergraph whose nodes are EC officials and stakeholders, and hyperedges connect entities that participate in the same meetings. Our analysis highlights the presence of a hierarchical core-periphery structure, with a few well-connected entities that occupy the center of the network and enjoy a stable integration in the EC policy-making process. Examination of the core composition reveals that companies and trade associations maintain closer relationships with the EC. A regression analysis shows that lobbying efforts and company size are significant predictors of company centrality, independent of other objective characteristics. Our findings provide quantitative evidence supporting the perception of lobbying as a tool dominated by well-connected actors, while also revealing heterogeneous lobbying strategies across stakeholder groups.
- Ferdinand Le Coz (IRIT, Toulouse)
Title: Unsupervised detection of persistent communities in dynamic networks using Networks embeddings.
Abstract: My research focuses on the use of Graph Neural Networks for the study of dynamic networks. I propose a general framework to discover dynamic communities without a-priori knowledge of these communities. I will present a synthetic dynamic graph and features generator to investigate the performances of the proposed framework. I will further discuss the impact of features on the representations produced by GNNs. Finally, I will test the method on a real-world Twitter dataset of retweets to try to recover the political communities of users.
- Gaspard Abel (ENS Paris-Saclay, Centre Borelli & CAMS)
Title: Uncovering social network activity using joint Temporal Point Processes for user and topic interaction
Abstract: Online social platforms have drastically affected the way people apprehend the information flows to which they are exposed, and the spread of multiple pieces of information or beliefs in a networked population is rarely uncorrelated. In this work we introduce here the Mixture of Interacting Cascades (MIC) model, which is based on marked Multidimensional Hawkes processes. MIC models jointly non-trivial interactions between users and information cascades. Moreover, we show how derivative layered visualizations can reveal the coupling between social network structure and topic alignments.
Lien d’evenement : https://cams.ehess.fr/diffusion-dopinions-desinformation
Well, the moment has come for our research team, in Centre Borelli ENS Paris-Saclay, currently known Machine Learning and Massive Data Analysis (MLMDA), to rethink the way we present ourselves, with a name the expresses well our current research affaires.
And the new name is … [drum roll]
Learning and Information Processing Systems (LIPS)
It comprises two parts:
- “Learning Systems”: This corresponds to all the challenges behind machine learning: mathematical modeling, learning theory, optimization, statistics, etc.
- “Information Processing Systems”: This part expresses the broad range of data types and applications we work on.
As for the term “Systems”, it highlights the systemic view we take when looking over complex problems (e.g. applications, industrial problems, etc.), which may involve the know-how from several fields and subfields.
LIPS is supposed to replace the previous MLMDA name, but its use will remain mainly internal for organizational purposes (for instance, our seminars will be called from now on the LIPS seminars). This is a choice that has been made all these years, to present ourselves publicly less as a distinct research team and rather as a team that is part of a very dynamic research center… the Centre Borelli, where teams and persons are in regular exchange and cross the borders of their main field(s) of interest as part of their unique way of doing things.
The Foundation FMJH has announced the call for post-doc funding 2024. Check a subject proposal entitled Predictive maintenance for networked systems. Applications are due to the 9th of December [link].
Contact me for more details.
Check also the other subject coming from the MLMDA group or Centre Borelli.
Here are some possibilities to define a project as part of the ARIA internship 2023-2024. For each of them, there is the possibility to continue further studying the subject in collaboration with a lab in France or abroad.
Topics:
- Graph theoretical Machine Learning
- Data clustering
- Statistical testing in high dimensions
- Interpretable/Explainable Machine Learning
- Social network analysis and diffusion of information
- Problems related to sequential decision-making
- Problems related to epidemic spreading and control
- Problems about urbanism and mobility (e.g. transportation, public safety, etc)
This event is an excellent opportunity to refresh my memory about the wonderful time -just over a decade- I spent in Ioannina, while studying at the University. I am looking forward to seeing again friends and familiar faces, and above all chat with the other colleagues and the young students of the department.
Many thanks to Evaggelia Pitoura for the invitation, and of course for organizing such events at the Computer Science and Engineering Department.
The Foundation FMJH has announced the call for post-doc funding 2023. Check our subject proposal entitled Graph operator pursuit for efficient graph machine learning. Applications are due to the 1st of December.
Contact me for more details on that.
Check also the other subject coming from the MLMDA group, which is related to Physics informed Neural Networks.
In the paper The Grey Hoodie Project: Big Tobacco, Big Tech, and the Threat on Academic Integrity, the authors investigate the very important issue of how the AI scientific agenda and policymaking around AI can be (and evidence imply they actually are) manipulated by Big Tech.
The abstract states:
“…Big Tech can actively distort the academic landscape to suit its needs. By comparing the well-studied actions of another industry (Big Tobacco) to the current actions of Big Tech we see similar strategies employed by both industries. These strategies enable either industry to sway and influence academic and public discourse. We examine the funding of academic research as a tool used by Big Tech to put forward a socially responsible public image, influence events hosted by and decisions made by funded universities, influence the research questions and plans of individual scientists, and discover receptive academics who can be leveraged…”
At the top of the post there is a figure showing how the different entities interact, and below the two main figures of the author’s study showing:
- Left: the percentage of Computer Science faculty of the sampled Ivy League Universities in US (UofT, MIT, Stanford, Berkeley) that have been funded by Big Tech (Google, Microsoft, Amazon, IBM, Facebook, Nvidia),
- Right: similarly for faculty affiliation.
The big questions are sort of obvious:
– How can Big Tech companies, which are among the most natural subjects of societal and legal monitoring/regulation, be the “pioneers” and “primary investigators” of the underlying ethics or democratic principles that those regulations should serve?
– How can Academia play its role to demand fair and equitable development, when part of it becomes a vehicle of legitimation for the big private interests?
The Franco-Bavarian AI Cup is now open for participation, for students and junior researchers from Europe interested in AI and Data Science.
Trophy: up to €95.000 for financing the launch of a start-up.
Procedure: Successful participation in one of the announced challenging use-cases, linked with sustainable development, which are out in collaboration with our partners from France and Germany:
- challenge from E.ON related to the energy distribution;
- challenge from Trading Hub Europe related to the pricing of energy;
- challenge from Deutsche Bahn Regio on the optimization of the planning of regional bus lines
- challenge from Centre Borelli related to the frailty of the elderly
- challenge from CEA on the detection of seismic activity.
You are invited to pass this information to your network.
Final registration date… this Friday!
More info: https://www.uni-passau.de/ai-