I have a Computer Science background, and I work on Machine Learning, or more specifically at the intersection between Computer Science and Applied Mathematics.

Short bio: ...

• Publications
• Long CV (last update: 23 March 2025)

Currently

Position: Permanent senior researcher at the Centre Giovanni Borelli¹, Ecole Normale Supérieure Paris-Saclay (ENS Paris-Saclay), University Paris-Saclay, France, and member of the Learning and Information Processing Systems (LIPS) research group². ...

Research focus: Coordinator of the ML on Graphs research theme.

Research initiatives: Member of the Industrial Data Analytics and Machine Learning Chair, having as industrial partners SNCF, CEA, Michelin. Other Academia-Industry synergies: join probject with RTE&SNCF, and collaboration with EDF. — Other initiatives: Since 2017, (co-)organizer of the LIPS seminars.

Past funding: ...

Current PhD students:

• Gaëtan Serré (~ global optimization)
• Gwendal Debaussart (~ graph representation learning)
• Eloi Campagne (~ time-series forecasting for the energy sector)
• Gaspard Abel (~ machine learning on social interaction networks)
• Gabriel Singer (~ statistical machine learning in uncertainty (imperfect labels))
• Jean Velluet (~ social network analysis on natural resources exploitation (fishery))

Alumni: ...

Announcements

News [bits and pieces]

• ΤCSS: The article Uncovering Social Network Activity Using Joint User and Topic Interaction, co-authored with G. Abel, J.-P. Nadal, and J. randon-Furling, has been accepted to the journal IEEE Transactions on Computational Social Systems [pdf][code].
• Preprints:
  • Enhancing Exploration in Global Optimization by Noise Injection in the Probability Measures Space, co-authored with G. Serré, P. Germain, and S. Gruffaz, 2026. [pdf]
  • Parametrized Power-Iteration Clustering for Directed Graphs, co-authored with G. Debaussart-Joniec, H. Sevi, and M. Jonckheere, 2026. [pdf]
  • Cascaded Transfer: Learning Many Tasks under Budget Constraints, co-authored with E. Campagne, Y. Amara-Ouali, Y. Goude, and M. Mougeot, 2026. [pdf]
  • Optimal Fair Aggregation of Crowdsourced Noisy Labels using Demographic Parity Constraints, co-authored with S. Gruffaz, G. Singer, O. Vo Van, and N. Vayatis, 2026. [pdf soon]
  • Reducing Recurrent Competitive Epidemics via Dynamic Resource Allocation, co-authored with G. Abel, and S. Sarao-Mannelli, 2025. [pdf]
  • Multi-view diffusion geometry using intertwined diffusion trajectories, co-authored with G. Debaussart-Joniec, 2025. [pdf][code].
  • Formal equivalence between global optimization consistency and random search, co-authored with G. Serré and N. Vayatis, 2025. [pdf][Lean code].
• ARCOM conf.: Gaspard Abel presented our work on multivariate Hawkes processes for social interactions and visualization at ARCOM’s event (co-organized with ENS P.S.). His talk had the title Analyse multi-échelle de la diffusion d’informations dans les réseaux sociaux [read more].
• PhD position: We are actively looking to fill an open PhD position on online change-point detection for structured multivariate time series. Read more.
• Preprint: the new preprint Graph Neural Networks for Electricity Load Forecasting consolidates two works that has been earlier presented in two ECML Workshops (2024 & 2025). Co-authored with E. Campagne, Y. Amara-Ouali, Y. Goude, and I. Zehavi. [pdf][code].
• Code: The GraphToolbox is a Python package designed for graph machine learning focused on time-series forecasting. It provides tools for data handling, model building, training, evaluation, and visualization. Developed as part of the work of our PhD student Eloi Campagne.
• JMLR: The paper Collaborative likelihood-ratio estimation over graphs, co-authored with A. de la Concha and N. Vayatis has been accepted in the Journal of Machine Learning Research. [pdf][code]
• Preprint: a new version of the working paper Generalized Dirichlet Energy and Graph Laplacians for Clustering Directed and Undirected Graphs, in which we present the a Generalized Spectral Clustering methodology. Except from H. Sevi and M. Jonckheere with who we initiated this project, now G. Debaussart-Joniec and M. Hacini have joined and helped with the “refurnishing” and thorough empirical validation [pdf].
• New pre-doc: Gabriel Singer has started a pre-doc with us and in collaboration with SNCF (in the frame of the IDAML Chair).
• New intern:
  • Mohamed Ben Loughmari (M1 ENSTA Paris) on advanced learning techniques for temporal point processes (Oct–Dec).
• Pattern Recognition: The paper UniForCE: The unimodality forest method for clustering and estimation of the number of clusters, co-authored with G. Vardakas and A. Likas, has been accepted to the Pattern Recognition journal (Elsevier) and will appear soon. [pdf to be updated][code].
• ECML 2025: the paper Plugging Attention Matrices into Power Grids: Towards Transparent Forecasting, co-authored along with E. Campagne, I. Zehavi, Y. Amara-Ouali, and Y. Goude, will appear at the ECML 2025 Workshop on Machine Learning for Sustainable Energy Systems (ML4SPS). [pdf]
• ANR oroject: The project SCOPED has been accepted by the ANR (the french national research agency). We are excited for the excellent potential offered by this project for exchanges and collaborations. The project is led by Cédric Richard, Laboratoire Lagrange, Université Côte d’Azur, and from Centre Borelli there will be involvement by L. Oudre, N. Vayatis, and myself.
• Code: Check the MultiLayerNetViz tool, which is developed in our team by Gaspard Abel, for visualization of social interaction data in a multilayer network layout.
• Preprint: a new version of the working paper Efficient stream-based Max-Min diversification with minimal failure rate has been posted on arxiv [pdf]. Initially a project initiated during Mathilde Fekom’s PhD, advances now after the M2 intern Yutai Nazir Zhao has joined forces.
• Preprint: the preprint Uncovering Social Network Activity Using Joint User and Topic Interaction has been posted online. It has been co-authored with Gaspard Abel, Jean-Pierre Nadal, and Julien Randon-Furling [pdf].
• Code: Excellent initiative by our PhD student Gaëtan Serré to release GOB: Global Optimization Benchmark, a collection of global optimization algorithms implemented in C++ and linked with Python. It also includes a set of analytical benchmark functions and a random function generator (PyGKLS) to test the performance of these algorithms.
• EUSIPCO 2025: The paper Online centralized non-parametric change-point detection via graph-based likelihood-ratio estimation has been accepted to the European Signal Processing Conference (EUSIPCO) 2025. Co-authored with A. de la Concha and N. Vayatis [pdf][long version][slides].
• PhD defense: on the 19th May, 2025, Alejandro de la Concha has successfully defended his PhD thesis on “Graph-based machine learning for detection tasks on complex systems“. Bravo Alex!! Proud for him and his excellent work.
• JdS 2025: Three works from our team will be presented at the Journée de Statistique 2025 (french conference of the Statistics community),
  • Fusing multiple data views with diffusion maps. G. Debaussart and A. Kalogeratos.
  • Graph-based weight cascading methods for multisite time series forecasting. E. Campagne at al.
  • Identification of structure features of graphs to improve GNN performance. Itai Zehavi et al.

See all news...