EN FR
EN FR
MAGNET - 2025

2025‌Activity reportProject-TeamMAGNET‌​‌

RNSR: 201321079K
  • Research center​​ Inria Centre at the​​​‌ University of Lille
  • In‌ partnership with:CNRS, Université‌​‌ de Lille
  • Team name:​​ Machine Learning in Information​​​‌ Networks
  • In collaboration with:‌Centre de Recherche en‌​‌ Informatique, Signal et Automatique​​ de Lille

Creation of​​​‌ the Project-Team: 2016 May‌ 01

Each year, Inria‌​‌ research teams publish an​​ Activity Report presenting their​​​‌ work and results over‌ the reporting period. These‌​‌ reports follow a common​​ structure, with some optional​​​‌ sections depending on the‌ specific team. They typically‌​‌ begin by outlining the​​ overall objectives and research​​​‌ programme, including the main‌ research themes, goals, and‌​‌ methodological approaches. They also​​​‌ describe the application domains​ targeted by the team,​‌ highlighting the scientific or​​ societal contexts in which​​​‌ their work is situated.​

The reports then present​‌ the highlights of the​​ year, covering major scientific​​​‌ achievements, software developments, or​ teaching contributions. When relevant,​‌ they include sections on​​ software, platforms, and open​​​‌ data, detailing the tools​ developed and how they​‌ are shared. A substantial​​ part is dedicated to​​​‌ new results, where scientific​ contributions are described in​‌ detail, often with subsections​​ specifying participants and associated​​​‌ keywords.

Finally, the Activity​ Report addresses funding, contracts,​‌ partnerships, and collaborations at​​ various levels, from industrial​​​‌ agreements to international cooperations.​ It also covers dissemination​‌ and teaching activities, such​​ as participation in scientific​​​‌ events, outreach, and supervision.​ The document concludes with​‌ a presentation of scientific​​ production, including major publications​​​‌ and those produced during​ the year.

Keywords

Computer​‌ Science and Digital Science​​

  • A3.1.11. Structured data
  • A3.4.​​​‌ Machine learning and statistics​
  • A5.7.2. Music
  • A5.7.3. Speech​‌
  • A5.8. Natural language processing​​
  • A9.1. Knowledge
  • A9.2. Machine​​​‌ learning
  • A9.4. Natural language​ processing
  • A9.7. AI algorithmics​‌
  • A9.8. Reasoning
  • A9.9. Distributed​​ AI, Multi-agent
  • A9.10. Hybrid​​​‌ approaches for AI
  • A9.11.​ Generative AI
  • A9.13. Agentic​‌ AI
  • A9.14. Evaluation of​​ AI models
  • A9.17. Cybersecurity​​​‌ and AI

Other Research​ Topics and Application Domains​‌

  • B2. Digital health
  • B9.5.1.​​ Computer science
  • B9.5.6. Data​​​‌ science
  • B9.6.1. Psychology
  • B9.6.8.​ Linguistics
  • B9.6.10. Digital humanities​‌
  • B9.9. Ethics
  • B9.10. Privacy​​

1 Team members, visitors,​​​‌ external collaborators

Research Scientists​

  • Pascal Denis [INRIA​‌, Researcher]
  • Raouf​​ Kerkouche [INRIA,​​​‌ Researcher, from Oct​ 2025]
  • Batiste Le​‌ Bars [INRIA,​​ ISFP]
  • Michael Perrot​​​‌ [INRIA, ISFP​]
  • Jan Ramon [​‌INRIA, Senior Researcher​​, HDR]
  • Damien​​​‌ Sileo [INRIA,​ ISFP]

Faculty Members​‌

  • Marc Tommasi [Team​​ leader, UNIV LILLE​​​‌, Professor, HDR​]
  • Angèle Brunelliere [​‌UNIV LILLE, Professor​​ Delegation, until Aug​​​‌ 2025]
  • Remi Gilleron​ [UNIV. LILLE,​‌ Emeritus, HDR]​​
  • Mikaela Keller [UNIV​​​‌ LILLE, Associate Professor​, from Sep 2025​‌]
  • Mikaela Keller [​​UNIV LILLE, Associate​​​‌ Professor Delegation, until​ Aug 2025]

Post-Doctoral​‌ Fellows

  • Arnaud Descours [​​INRIA, Post-Doctoral Fellow​​​‌, until Aug 2025​]
  • Luis Eduardo Lugo​‌ Martinez [INRIA,​​ Post-Doctoral Fellow]

PhD​​​‌ Students

  • Paul Andrey [​INRIA]
  • Antoine Barczewski​‌ [INRIA, until​​ Oct 2025]
  • Mouad​​​‌ Blej [INRIA]​
  • Nassim Boudjenah [INRIA​‌, from Dec 2025​​]
  • Edwige Cyffers [​​​‌UNIV LILLE, until​ Mar 2025]
  • Marc​‌ Damie [UNIV. TWENTE​​]
  • Jean Dufraiche [​​​‌INRIA]
  • Brahim Erraji​ [INRIA]
  • Dimitri​‌ Kachler [INRIA,​​ from Nov 2025]​​​‌
  • Aleksei Korneev [UNIV​ LILLE]
  • Dinh-Viet-Toan Le​‌ [UNIV LILLE,​​ until Oct 2025]​​​‌
  • Bastien Lietard [INRIA​]
  • Gabriel Loiseau [​‌HORNET SECURITY, CIFRE​​]
  • Aymane Moataz [​​​‌INRIA, until Jun​ 2025]
  • Clement Pierquin​‌ [CRAFT.AI, CIFRE​​]
  • Aurelien Said Housseini​​ [INRIA, until​​​‌ May 2025]
  • Quentin‌ Sinh [INRIA]‌​‌
  • Shreya Venugopal [INRIA​​]

Technical Staff

  • Amer​​​‌ Alzein [INRIA,‌ Engineer, from Jun‌​‌ 2025 until Sep 2025​​]
  • Jules Boulet [​​​‌INRIA, Engineer,‌ until May 2025]‌​‌
  • Baptiste Cottier [INRIA​​, Engineer]
  • Simon​​​‌ Decomble [INRIA,‌ Engineer, from Apr‌​‌ 2025]
  • Leonard Deroose​​ [INRIA, Engineer​​​‌]
  • Zakaria El Bouchouari‌ [INRIA, Engineer‌​‌, from Sep 2025​​]
  • Younes Ikli [​​​‌INRIA, Engineer]‌
  • Valentin Lacombe [INRIA‌​‌, Engineer, from​​ Feb 2025]
  • Alexandre​​​‌ Louvet [INRIA,‌ Engineer, from Sep‌​‌ 2025]
  • Victor Roussanaly​​ [INRIA, Engineer​​​‌, from Sep 2025‌]
  • Elina Thibeau-Sutre [‌​‌INRIA, Engineer,​​ until Jun 2025]​​​‌
  • Jules Yvon [INRIA‌, Engineer]

Interns‌​‌ and Apprentices

  • Thomas Bobille​​ [INRIA, Intern​​​‌, from Apr 2025‌ until Aug 2025]‌​‌
  • Yassine Oj [INRIA​​, Intern, from​​​‌ Apr 2025 until May‌ 2025]
  • Valentin Quesnel-Dumont‌​‌ [INRIA, Intern​​, from Apr 2025​​​‌ until Aug 2025]‌
  • Mohamed El Amine Serradj‌​‌ [INRIA, Intern​​, from Apr 2025​​​‌ until Aug 2025]‌

Administrative Assistants

  • Nathalie Bonte‌​‌ [INRIA, from​​ Jun 2025]
  • Aurore​​​‌ Dalle [INRIA,‌ until May 2025]‌​‌

2 Overall objectives

The​​ main objective of Magnet​​​‌ is to develop original‌ machine learning methods for‌​‌ networked data. We consider​​ information networks in which​​​‌ the data consist of‌ feature vectors or texts.‌​‌ We model such networks​​ as graphs wherein nodes​​​‌ correspond to entities (documents,‌ spans of text, users,‌​‌ datasets, learners etc.) and​​ edges correspond to relations​​​‌ between entities (similarity, answer,‌ co-authoring, friendship etc.). In‌​‌ Mining and Learning in​​ Graphs, our main​​​‌ research goal is to‌ efficiently search for the‌​‌ best hidden graph structure​​ to be generated for​​​‌ solving a given learning‌ task which exploits the‌​‌ relationships between entities. In​​ Machine Learning for Natural​​​‌ Language Processing the objective‌ is to go beyond‌​‌ vectorial classification to solve​​ tasks like coreference resolution​​​‌ and entity linking, temporal‌ structure prediction, and discourse‌​‌ parsing. In Decentralized Machine​​ Learning we address the​​​‌ problem of learning in‌ a private, fair and‌​‌ energy efficient way when​​ data are naturally distributed​​​‌ in a network.

The‌ challenges are the dimensionality‌​‌ of the input space,​​ possibly the dimensionality of​​​‌ the output space, the‌ high level of dependencies‌​‌ between the data, the​​ inherent ambiguity of textual​​​‌ data and the limited‌ amount of human labeling.‌​‌ We are interested in​​ making machine learning approaches​​​‌ more acceptable to society.‌ Privacy, sobriety and fairness‌​‌ are important issues that​​ pertain to this research​​​‌ line, and we are‌ interested in the empowerment‌​‌ of end users in​​ the machine learning processes.​​​‌

3 Research program

The‌ research program of Magnet‌​‌ is structured along three​​ main axes.

  • Axis 1:​​​‌ Mining and Learning in‌ Graphs

    This axis is‌​‌ the backbone of the​​​‌ team. Most of the​ techniques and algorithms developed​‌ in this axis are​​ known by the team​​​‌ members and have impact​ on the two other​‌ axes. We address the​​ following questions and objectives:​​​‌

    How to adaptively build​ graphs with respect to​‌ the given tasks? We​​ study adaptive graph construction​​​‌ along several directions. The​ first one is to​‌ learn the best similarity​​ measure for the graph​​​‌ construction. The second one​ is to combine different​‌ views over the data​​ in the graph construction​​​‌ and learn good representations.​ We also study weak​‌ forms of supervision like​​ comparisons.

    How to design​​​‌ methods able to achieve​ a good trade-off between​‌ predictive accuracy and computational​​ complexity? We develop new​​​‌ algorithms for efficient graph-based​ learning (for instance node​‌ prediction or link prediction).​​ In order to deal​​​‌ with scalability issues, our​ approach is based on​‌ optimization, graph sparsification techniques​​ and graph sampling methods.​​​‌

    How to find patterns​ in graphs based on​‌ efficient computations of some​​ statistics? We develop graph​​​‌ mining algorithms and statistics​ in the context of​‌ correlated data.

  • Axis 2:​​ Machine Learning for Natural​​​‌ Language Processing
    In this​ axis, we address the​‌ general question that relates​​ graph-based learning and Natural​​​‌ Language Processing (NLP): How​ to go beyond vectorial​‌ classification models in NLP​​ tasks? We study the​​​‌ combination of learning representation,​ structured prediction and graph-based​‌ learning methods. Data sobriety​​ and fairness are major​​​‌ constraints we want to​ deal with. The targeted​‌ NLP tasks are coreference​​ resolution and entity linking,​​​‌ temporal structure prediction, and​ discourse parsing.
  • Axis 3:​‌ Decentralized Machine Learning and​​ Privacy
    In this axis,​​​‌ we study How to​ design private by design​‌ machine learning algorithms? Taking​​ as an opportunity the​​​‌ fact that data collection​ is now decentralized on​‌ smart devices, we propose​​ alternatives to large data​​​‌ centers where data are​ gathered by developing collaborative​‌ and personalized learning.

Contrary​​ to many machine learning​​​‌ approaches where data points​ and tasks are considered​‌ in isolation, we think​​ that a key point​​​‌ of this research is​ to be able to​‌ leverage the relationships between​​ data and learning objectives.​​​‌ Therefore, using graphs as​ an abstraction of information​‌ networks is a major​​ playground for Magnet.​​​‌ Research related to graph​ data is a transversal​‌ axis, describing a layer​​ of work supporting two​​​‌ other axes on Natural​ Language Processing and decentralized​‌ learning. The machine learning​​ and mining in graphs​​​‌ communities have evolved, for​ instance taking into account​‌ data streams, dynamics but​​ maybe more importantly, focusing​​​‌ on deep learning. Deep​ neural nets are here​‌ to stay, and they​​ are useful tools to​​​‌ tackle difficult problems so​ we embrace them at​‌ different places in the​​ three axes.

Magnet conducts​​​‌ research along the three​ axes described above but​‌ will put more emphasis​​ on social issues of​​​‌ machine learning. In the​ context of the recent​‌ deployment of artificial intelligence​​ into our daily lives,​​​‌ we are interested in​ making machine learning approaches​‌ more acceptable to society.​​ Privacy, sobriety and fairness​​ are important issues that​​​‌ pertain to this research‌ line, but more generally‌​‌ we are interested in​​ the empowerment of end​​​‌ users in the machine‌ learning processes. Reducing the‌​‌ need of one central​​ authority and pushing more​​​‌ the data processing on‌ the user side, that‌​‌ is decentralization, also participates​​ to this effort. Reducing​​​‌ resources means reducing costs‌ and energy and contributes‌​‌ to building more accessible​​ technologies for companies and​​​‌ users. By considering learning‌ tasks in a more‌​‌ personalized way, but increasing​​ collaboration, we think that​​​‌ we can design solutions‌ that work in low‌​‌ resources regime, with less​​ data or supervision.

In​​​‌ Magnet we emphasize a‌ different approach than blindly‌​‌ brute-forcing tasks with loads​​ of data. Applications to​​​‌ social sciences for instance‌ have different needs and‌​‌ constraints that motivate data​​ sobriety, fairness and privacy.​​​‌ We are interested in‌ weaker supervision, by leveraging‌​‌ structural properties described in​​ graphs of data, relying​​​‌ on transfer and multi-task‌ learning when faced with‌​‌ graphs of tasks and​​ users. Algorithmic and statistical​​​‌ challenges related to the‌ graph structure of the‌​‌ data still contain open​​ questions. On the statistical​​​‌ side, examples are to‌ take dependencies into account,‌​‌ for instance to compute​​ a mean, to reduce​​​‌ the need of sampling‌ by exploiting known correlations.‌​‌ For the algorithmic point​​ of view, going beyond​​​‌ unlabeled undirected graphs, in‌ particular considering attributed graphs‌​‌ containing text or other​​ information and addressing the​​​‌ case of distributed graphs‌ while maintaining formal guarantees‌​‌ are getting more attention.​​

In the second axis​​​‌ devoted to NLP, we‌ focus our research on‌​‌ graph-based and representation learning​​ into several directions, all​​​‌ aiming at learning richer,‌ more robust, and more‌​‌ transferable linguistic representations.​​ This research program will​​​‌ attempt to bring about‌ strong cross-fertilizations with the‌​‌ other axes, addressing problems​​ in graph, privacy and​​​‌ fairness and making links‌ with decentralized learning. At‌​‌ the intersection between graph-based​​ and representation learning, we​​​‌ will first develop graph‌ embedding algorithms for deriving‌​‌ linguistic representations which are​​ able to capture higher-level​​​‌ semantic and world-knowledge information‌ which eludes strictly distributional‌​‌ models. As an initial​​ step, we envision leveraging​​​‌ pre-existing ontologies (e.g., WordNet,‌ DBpedia), from which one‌​‌ can easily derive interesting​​ similarity graphs between words​​​‌ or noun phrases. We‌ also plan to investigate‌​‌ innovative ways of articulating​​ graph-based semi-supervised learning algorithms​​​‌ and word embedding techniques.‌ A second direction involves‌​‌ learning representations that are​​ more robust to bias,​​​‌ privacy attacks and adversarial‌ examples. Thus, we intend‌​‌ to leverage recent adversarial​​ training strategies, in which​​​‌ an adversary attempts to‌ recover sensitive attributes (e.g.,‌​‌ gender, race) from the​​ learned representations, to be​​​‌ able to neutralize bias‌ or to remove sensitive‌​‌ features. An application domain​​ for this line of​​​‌ research is for instance‌ speech data. The study‌​‌ of learning private representation​​ with its link to​​​‌ fairness in the decentralized‌ setting is another important‌​‌ research topic for the​​ team. In this context​​​‌ of fairness, we also‌ intend to develop similar‌​‌ algorithms for detecting slants,​​​‌ and ultimately for generating​ de-biased or “re-biased” versions​‌ of text embeddings. An​​ illustration is on political​​​‌ slant in written texts​ (e.g., political speeches and​‌ manifestos). Thirdly, we intend​​ to learn linguistic representations​​​‌ that can transfer more​ easily across languages and​‌ domains, in particular in​​ the context of structured​​​‌ prediction problems for low-resource​ languages. For instance, we​‌ first propose to jointly​​ learn model parameters for​​​‌ each language (and/or domains)​ in a multi-task setting,​‌ and leverage a (pre-existing​​ or learned) graph encoding​​​‌ structural similarities between languages​ (and/or domains). This type​‌ of approach would nicely​​ tie in with our​​​‌ previous work on multilingual​ dependency parsing and on​‌ learning personalized models. Furthermore,​​ we will also study​​​‌ how to combine and​ adapt some neural architectures​‌ recently introduced for sequence-to-sequence​​ problems in order to​​​‌ enable transfer of language​ representations.

In terms of​‌ technological transfer, we maintain​​ collaborations with researchers in​​​‌ the humanities and the​ social sciences, helping them​‌ to leverage state-of-the-art NLP​​ techniques to develop new​​​‌ insights to their research​ by extracting relevant information​‌ from large amounts of​​ texts.

The third axis​​​‌ is on distributed and​ decentralized learning and privacy​‌ preserving machine learning. Recent​​ years have seen the​​​‌ evolution of information systems​ towards ubiquitous computing, smart​‌ objects and applications fueled​​ by artificial intelligence. Data​​​‌ are collected on smart​ devices like smartphones, watches,​‌ home devices etc. They​​ include texts, locations, social​​​‌ relationships. Many sensitive data​ —race, gender, health conditions,​‌ tastes etc— can be​​ inferred. Others are just​​​‌ recorded like activities, social​ relationships but also biometric​‌ data like voice and​​ measurements from sensor data.​​​‌ The main tendency is​ to transfer data into​‌ central servers mostly owned​​ by a few tier​​​‌ parties. The situation generates​ high privacy risks for​‌ the users for many​​ reasons: loss of data​​​‌ control, unique entry point​ for data access, unsolicited​‌ data usage etc. But​​ it also increases monopolistic​​​‌ situations and tends to​ develop oversized infrastructures. The​‌ centralized paradigm also has​​ limits when data are​​​‌ too huge such as​ in the case of​‌ multiple videos and sensor​​ data collected for autonomous​​​‌ driving. Partially or fully​ decentralized systems provide an​‌ alternative, to emphasis data​​ exploitation rather than data​​​‌ sharing. For Magnet,​ they are source of​‌ many new research directions​​ in machine learning at​​​‌ two scales: at the​ algorithmic level and at​‌ a systemic level.

At​​ the algorithmic level the​​​‌ question is to develop​ new privacy preserving algorithms​‌ in the context of​​ decentralized systems. In this​​​‌ context, data remains where​ it has been collected​‌ and learning or statistical​​ queries are processed at​​​‌ the local level. An​ important question we study​‌ is to take into​​ account and measure the​​​‌ impact of collaboration. We​ also aim at developing​‌ methods in the online​​ setting where data arrives​​​‌ continuously or participants join​ and leave the collaboration​‌ network. The granularity of​​ exchanges, the communication cost​​​‌ and the dynamic scenarios,​ are also studied. On​‌ the privacy side, decentralization​​ is not sufficient to​​ establish privacy guarantees because​​​‌ learned models together with‌ the dynamics of collaborative‌​‌ learning may reveal private​​ training data if the​​​‌ models are published or‌ if the communications are‌​‌ observed. But, although it​​ has not been yet​​​‌ well established, decentralization can‌ naturally increase privacy-utility ratio.‌​‌ A direction of research​​ is to formally prove​​​‌ the privacy gain when‌ randomized decentralized protocols are‌​‌ used during learning. In​​ some situations, for instance​​​‌ when part of the‌ data is not sensitive‌​‌ or when trusted servers​​ can be used, a​​​‌ combination between a fully‌ decentralized and a centralized‌​‌ approach is very relevant.​​ In this setting, the​​​‌ question is to find‌ a good trade-off between‌​‌ local versus global computations.​​

At the systemic layer,​​​‌ in Magnet we feel‌ that there is a‌​‌ need for research on​​ a global and holistic​​​‌ level, that is to‌ consider full processes involving‌​‌ learning, interacting, predicting, reasoning,​​ repeating etc. rather than​​​‌ studying the privacy of‌ isolated learning algorithms. Our‌​‌ objective is to design​​ languages for describing processes​​​‌ (workflows), data (database schema,‌ background knowledge), population statistics,‌​‌ privacy properties of algorithms,​​ privacy requirements and other​​​‌ relevant information. This is‌ fully aligned with recent‌​‌ trends that aim at​​ giving to statistical learning​​​‌ a more higher level‌ of formal specifications and‌​‌ illustrates our objective for​​ more acceptable and transparent​​​‌ machine learning. We also‌ work towards more robust‌​‌ privacy-friendly systems, being able​​ to handle a wider​​​‌ range of malicious behavior‌ such as collusion to‌​‌ obtain information or inputting​​ incorrect data to obtain​​​‌ information or to influence‌ the result of collaborative‌​‌ computations. From the transfer​​ point of view, we​​​‌ plan to apply transparent,‌ privacy-friendly machine learning in‌​‌ significant application domains, such​​ as medicine, surveying, demand​​​‌ prediction and recommendation. In‌ this context, we are‌​‌ interested to understand the​​ appreciation of humans of​​​‌ transparency, verifiability, fairness, privacy-preserving‌ and other trust-increasing aspects‌​‌ of our technologies.

4​​ Application domains

Our application​​​‌ domains cover health, mobility,‌ social sciences and voice‌​‌ technologies.

  • Health
    Privacy is​​ of major importance in​​​‌ the health domain. We‌ contribute to develop methods‌​‌ to give access to​​ the use of data​​​‌ in a private way‌ rather than to the‌​‌ data itself centralized in​​ vulnerable single locations. As​​​‌ an example, we are‌ working with hospitals to‌​‌ develop the means of​​ multicentric studies with privacy​​​‌ guarantees. A second example‌ is personalized medicine where‌​‌ personal devices collect private​​ and highly sensitive data.​​​‌ Potential applications of our‌ research allow to keep‌​‌ data on device and​​ to privately compute statistics.​​​‌
  • Social sciences
    Our NLP‌ research activities are rooted‌​‌ in linguistics, but learning​​ unbiased representations of texts​​​‌ for instance or simply‌ identifying unfair representations also‌​‌ have impacts in political​​ sciences and history.
  • Music​​​‌ information retrieval
    By using‌ analogies between language and‌​‌ music (symbolic notation) we​​ tackle music information retrieval​​​‌ tasks such as style‌ classification and structure detection.‌​‌
  • Voice technologies
    We develop​​ methods for privacy in​​​‌ speech that can be‌ embedded in software suites‌​‌ dedicated to voice-based interaction​​​‌ systems.

5 Social and​ environmental responsibility

5.1 Footprint​‌ of research activities

Some​​ of our research activities​​​‌ are energy intensive and​ we will work to​‌ reduce this carbon footprint​​ in the future. Parts​​​‌ of the research projects​ FedMalin (see Section 10.3.2​‌) and FLUTE are​​ dedicated to this objective​​​‌ for the Federated Learning​ setting. In a collaboration​‌ with the Spirals team,​​ we have extended the​​​‌ DecLearn API with features​ that are dedicated to​‌ energy consumption measurement with​​ the PowerAPI library. We​​​‌ are working on designing​ active strategies to select​‌ and schedule client participation​​ in Federated learning, based​​​‌ on their energy consumption.​ The objective is to​‌ better handle the trade-off​​ between energy consumption and​​​‌ accuracy in settings where​ the energy budget is​‌ limited.

5.2 Impact of​​ research results

The main​​​‌ research topics of the​ team contribute to improve​‌ transparency, fairness and privacy​​ in machine learning and​​​‌ reduce bias in natural​ language processing.

6 Highlights​‌ of the year

  • Nicolas​​ Papernot now holds INRIA​​​‌ international chair visiting Premedical,​ Magnet and Privatics.
  • Two​‌ projects have been accepted​​ for a pluridisciplinary research:​​​‌ CDP Loop and Prime​ Next-Gen. They will support​‌ the application of our​​ research to the health​​​‌ domain.

7 Latest software​ developments, platforms, open data​‌

7.1 Latest software developments​​

7.1.1 CoRTeX

  • Name:
    Python​​​‌ library for noun phrase​ COreference Resolution in natural​‌ language TEXts
  • Keyword:
    Natural​​ language processing
  • Functional Description:​​​‌
    CoRTex is a LGPL-licensed​ Python library for Noun​‌ Phrase coreference resolution in​​ natural language texts. This​​​‌ library contains implementations of​ various state-of-the-art coreference resolution​‌ algorithms, including those developed​​ in our research. In​​​‌ addition, it provides a​ set of APIs and​‌ utilities for text pre-processing,​​ reading the CONLL2012 and​​​‌ CONLLU annotation formats, and​ performing evaluation, notably based​‌ on the main evaluation​​ metrics (MUC, B-CUBED, and​​​‌ CEAF). As such, CoRTex​ provides benchmarks for researchers​‌ working on coreference resolution,​​ but it is also​​​‌ of interest for developers​ who want to integrate​‌ a coreference resolution within​​ a larger platform. It​​​‌ currently supports use of​ the English or French​‌ language.
  • Contact:
    Pascal Denis​​
  • Participant:
    Pascal Denis

7.1.2​​​‌ Mangoes

  • Name:
    MAgnet liNGuistic​ wOrd vEctorS
  • Functional Description:​‌

    Mangoes is a toolbox​​ for constructing and evaluating​​​‌ static and contextual token​ vector representations (aka embeddings).​‌ The main functionalities are:​​

    - Contextual embeddings: Access​​​‌ a large collection of​ pretrained transformer-based language models,​‌ Pre-train a BERT language​​ model on a corpus,​​​‌ Fine-tune a BERT language​ model for a number​‌ of extrinsic tasks, Extract​​ features/predictions from pretrained language​​​‌ models.

    - Static embeddings:​ Process textual data and​‌ compute vocabularies and co-occurrence​​ matrices. Input data should​​​‌ be raw text or​ annotated text, Compute static​‌ word embeddings with different​​ state-of-the art unsupervised methods,​​​‌ Propose statistical and intrinsic​ evaluation methods, as well​‌ as some visualization tools,​​ Generate context dependent embeddings​​​‌ from a pretrained language​ model.

    Future releases will​‌ include methods for injecting​​ lexical and semantic knowledge​​​‌ into token and multi-model​ embeddings, and interfaces into​‌ common external knowledge resources.​​

  • URL:
  • Contact:
    Nathalie​​ Vauquier

7.1.3 metric-learn

  • Keywords:​​​‌
    Machine learning, Python, Metric‌ learning
  • Functional Description:

    Distance‌​‌ metrics are widely used​​ in the machine learning​​​‌ literature. Traditionally, practicioners would‌ choose a standard distance‌​‌ metric (Euclidean, City-Block, Cosine,​​ etc.) using a priori​​​‌ knowledge of the domain.‌ Distance metric learning (or‌​‌ simply, metric learning) is​​ the sub-field of machine​​​‌ learning dedicated to automatically‌ constructing optimal distance metrics.‌​‌

    This package contains efficient​​ Python implementations of several​​​‌ popular metric learning algorithms.‌

  • URL:
  • Contact:
    Aurélien‌​‌ Bellet
  • Partner:
    Parietal

7.1.4​​ MyLocalInfo

  • Keywords:
    Privacy, Machine​​​‌ learning, Statistics
  • Functional Description:‌
    Decentralized algorithms for machine‌​‌ learning and inference tasks​​ which (1) perform as​​​‌ much computation as possible‌ locally and (2) ensure‌​‌ privacy and security by​​ avoiding that personal data​​​‌ leaves devices.
  • Contact:
    Nathalie‌ Vauquier

7.1.5 declearn

  • Keyword:‌​‌
    Federated learning
  • Scientific Description:​​

    declearn is a python​​​‌ package providing with a‌ framework to perform federated‌​‌ learning, i.e. to train​​ machine learning models by​​​‌ distributing computations across a‌ set of data owners‌​‌ that, consequently, only have​​ to share aggregated information​​​‌ (rather than individual data‌ samples) with an orchestrating‌​‌ server (and, by extension,​​ with each other).

    The​​​‌ aim of declearn is‌ to provide both real-world‌​‌ end-users and algorithm researchers​​ with a modular and​​​‌ extensible framework that:

    (1)‌ builds on abstractions general‌​‌ enough to write backbone​​ algorithmic code agnostic to​​​‌ the actual computation framework,‌ statistical model details or‌​‌ network communications setup

    (2)​​ designs modular and combinable​​​‌ objects, so that algorithmic‌ features, and more generally‌​‌ any specific implementation of​​ a component (the model,​​​‌ network protocol, client or‌ server optimizer...) may easily‌​‌ be plugged into the​​ main federated learning process​​​‌ - enabling users to‌ experiment with configurations that‌​‌ intersect unitary features

    (3)​​ provides with functioning tools​​​‌ that may be used‌ out-of-the-box to set up‌​‌ federated learning tasks using​​ some popular computation frameworks​​​‌ (scikit- learn, tensorflow, pytorch...)‌ and federated learning algorithms‌​‌ (FedAvg, Scaffold, FedYogi...)

    (4)​​ provides with tools that​​​‌ enable extending the support‌ of existing tools and‌​‌ APIs to custom functions​​ and classes without having​​​‌ to hack into the‌ source code, merely adding‌​‌ new features (tensor libraries,​​ model classes, optimization plug-ins,​​​‌ orchestration algorithms, communication protocols...)‌ to the party.

    Parts‌​‌ of the declearn code​​ (Optimizers,...) are included in​​​‌ the FedBioMed software.

    At‌ the moment, declearn has‌​‌ been focused on so-called​​ "centralized" federated learning that​​​‌ implies a central server‌ orchestrating computations, but it‌​‌ might become more oriented​​ towards decentralized processes in​​​‌ the future, that remove‌ the use of a‌​‌ central agent.

  • Functional Description:​​

    This library provides the​​​‌ two main components to‌ perform federated learning:

    (1)‌​‌ the client, to be​​ run by each participant,​​​‌ performs the learning on‌ local data et releases‌​‌ only the result of​​ the computation

    (2) the​​​‌ server orchestrates the process‌ and aggregates the local‌​‌ models in a global​​ model

  • News of the​​​‌ Year:
    Two major releases‌ with key new functionalities‌​‌ including algorithms for group​​ fairness and the ability​​​‌ to use secure aggregation.‌
  • URL:
  • Contact:
    Aurélien‌​‌ Bellet
  • Participants:
    Paul Andrey,​​​‌ Aurélien Bellet, Nathan Bigaud,​ Marc Tommasi, Nathalie Vauquier​‌
  • Partner:
    CHRU Lille

7.1.6​​ fairgrad

  • Name:
    FairGrad: Fairness​​​‌ Aware Gradient Descent
  • Keywords:​
    Fairness, Fair and ethical​‌ machine learning, Machine learning,​​ Classification
  • Functional Description:
    FairGrad​​​‌ is an easy to​ use general purpose approach​‌ in Machine Learning to​​ enforce fairness in gradient​​​‌ descent based methods
  • URL:​
  • Contact:
    Michael Perrot​‌

7.1.7 tasksource

  • Name:
    tasksource​​
  • Keyword:
    Natural language processing​​​‌
  • Functional Description:
    tasksource streamlines​ interchangeable datasets usage to​‌ scale evaluation or multi-task​​ learning. All implemented preprocessings​​​‌ are in tasks.py or​ tasks.md. A preprocessing is​‌ a function that accepts​​ a dataset and returns​​​‌ the standardized dataset. Preprocessing​ code is concise and​‌ human-readable.
  • URL:
  • Publication:​​
  • Contact:
    Damien Sileo​​​‌

7.1.8 Voice Transformer 2​

  • Keywords:
    Speech, Privacy
  • Scientific​‌ Description:

    The implemented method​​ is inspired from the​​​‌ speaker anonymisation method proposed​ in [Fan+19], which performs​‌ voice conversion based on​​ x-vectors [Sny+18], a fixed-length​​​‌ representation of speech signals​ that form the basis​‌ of state-of-the-art speaker verification​​ systems. We have brought​​​‌ several improvements to this​ method such as pitch​‌ transformation, and new design​​ choices for x-vector selection​​​‌

    [Fan+19] F. Fang, X.​ Wang, J. Yamagishi, I.​‌ Echizen, M. Todisco, N.​​ Evans, and J.F. Bonastre.​​​‌ “Speaker Anonymization Using x-vector​ and Neural Waveform Models”.​‌ In: Proceedings of the​​ 10th ISCA Speech Synthesis​​​‌ Workshop. 2019, pp. 155–160.​ [Sny+18] D. Snyder, D.​‌ Garcia-Romero, G. Sell, D.​​ Povey, and S. Khudanpur.​​​‌ “X-vectors: Robust DNN embeddings​ for speaker recognition”. In:​‌ Proceedings of ICASSP 2018​​ - 2018 IEEE International​​​‌ Conference on Acoustics, Speech​ and Signal Processing (ICASSP).​‌ 2018, pp. 5329–5333.

  • Functional​​ Description:
    Voice Transformer increases​​​‌ the privacy of users​ of voice interfaces by​‌ converting their voice into​​ another person’s voice without​​​‌ modifying the spoken message.​ It ensures that any​‌ information extracted from the​​ transformed voice can hardly​​​‌ be traced back to​ the original speaker, as​‌ validated through state-of-the-art biometric​​ protocols, and it preserves​​​‌ the phonetic information required​ for human labelling and​‌ training of speech-to-text models.​​
  • Contact:
    Nathalie Vauquier
  • Participants:​​​‌
    Brij Mohan Lal Srivastava,​ Nathalie Vauquier, Emmanuel Vincent,​‌ Marc Tommasi

7.2 Open​​ data

Participants: Damien Sileo​​​‌, Valentin Lacombe,​ Valentin Quesnel.

Reasoning​‌ Core (rc0, rc1)
  • Contributors:​​
    Damien Sileo , Valentin​​​‌ Lacombe , Valentin Quesnel​
  • Description:
    Reasoning Core is​‌ a scalable environment for​​ Reinforcement Learning with Verifiable​​​‌ Rewards (RLVR) designed to​ advance foundational symbolic reasoning​‌ in LLMs. It procedurally​​ generates problems across core​​​‌ formal domains, including PDDL​ planning, first-order logic, context-free​‌ grammar parsing, causal reasoning,​​ and system equation solving.​​​‌ This release includes the​ generated datasets rc0 and​‌ rc1, designed respectively for​​ pre-training/mid-training (SFT) and post-training​​​‌ (RL).
  • Dataset PID (DOI,...):​
    arXiv:2509.18083
  • Project link:
  • Publications:
    V.​​ Lacombe, V. Quesnel, D.​​​‌ Sileo. Reasoning Core: A​ Scalable RL Environment for​‌ LLM Symbolic Reasoning. arXiv​​ preprint arXiv:2509.18083, 2025. 48​​​‌
  • Contact:
    Damien Sileo
  • Release​ contributions:
    Datasets and Code​‌ environment for LLM symbolic​​ reasoning

8 New results​​​‌

8.1 Natural Language Processing​

Participants: Damien Sileo,​‌ Bastien Liétard, Valentin​​ Lacombe, Angèle Brunellière​​.

Reasoning Core: A​​​‌ Scalable RL Environment for‌ LLM Symbolic Reasoning 48‌​‌

We introduce Reasoning Core,​​ a new scalable environment​​​‌ for Reinforcement Learning with‌ Verifiable Rewards (RLVR), designed‌​‌ to advance foundational symbolic​​ reasoning in Large Language​​​‌ Models (LLMs). Unlike existing‌ benchmarks that focus on‌​‌ games or isolated puzzles,​​ Reasoning Core procedurally generates​​​‌ problems across core formal‌ domains, including PDDL planning,‌​‌ first-order logic, context-free grammar​​ parsing, causal reasoning, and​​​‌ system equation solving. The‌ environment is built on‌​‌ key design principles of​​ high-generality problem distributions, verification​​​‌ via external tools, and‌ continuous difficulty control, which‌​‌ together provide a virtually​​ infinite supply of novel​​​‌ training instances. Initial zero-shot‌ evaluations with frontier LLMs‌​‌ confirm the difficulty of​​ Reasoning Core's tasks, positioning​​​‌ it as a promising‌ resource to improve the‌​‌ reasoning capabilities of future​​ models.

Bridging the Data​​​‌ Provenance Gap Across Text,‌ Speech and Video 33‌​‌

Progress in AI is​​ driven largely by the​​​‌ scale and quality of‌ training data. Despite this,‌​‌ there is a deficit​​ of empirical analysis examining​​​‌ the attributes of well-established‌ datasets beyond text. In‌​‌ this work we conduct​​ the largest and first-of-its-kind​​​‌ longitudinal audit across modalities–popular‌ text, speech, and video‌​‌ datasets–from their detailed sourcing​​ trends and use restrictions​​​‌ to their geographical and‌ linguistic representation. Our manual‌​‌ analysis covers nearly 4000​​ public datasets between 1990-2024,​​​‌ spanning 608 languages, 798‌ sources, 659 organizations, and‌​‌ 67 countries. We find​​ that multimodal machine learning​​​‌ applications have overwhelmingly turned‌ to web-crawled, synthetic, and‌​‌ social media platforms, such​​ as YouTube, for their​​​‌ training sets, eclipsing all‌ other sources since 2019.‌​‌ Secondly, tracing the chain​​ of dataset derivations we​​​‌ find that while less‌ than 33% of datasets‌​‌ are restrictively licensed, over​​ 80% of the source​​​‌ content in widely-used text,‌ speech, and video datasets,‌​‌ carry non-commercial restrictions. Finally,​​ counter to the rising​​​‌ number of languages and‌ geographies represented in public‌​‌ AI training datasets, our​​ audit demonstrates measures of​​​‌ relative geographical and multilingual‌ representation have failed to‌​‌ significantly improve their coverage​​ since 2013. We believe​​​‌ the breadth of our‌ audit enables us to‌​‌ empirically examine trends in​​ data sourcing, restrictions, and​​​‌ Western-centricity at an ecosystem-level,‌ and that visibility into‌​‌ these questions are essential​​ to progress in responsible​​​‌ AI. As a contribution‌ to ongoing improvements in‌​‌ dataset transparency and responsible​​ use, we release our​​​‌ entire multimodal audit, allowing‌ practitioners to trace data‌​‌ provenance across text, speech,​​ and video.

Humanity's Last​​​‌ Exam 23

Benchmarks are‌ important tools for tracking‌​‌ the rapid advancements in​​ large language model (LLM)​​​‌ capabilities. However, benchmarks are‌ not keeping pace in‌​‌ difficulty: LLMs now achieve​​ over 90% accuracy on​​​‌ popular benchmarks like MMLU,‌ limiting informed measurement of‌​‌ state-of-the-art LLM capabilities. In​​ response, we introduce Humanity's​​​‌ Last Exam (HLE), a‌ multi-modal benchmark at the‌​‌ frontier of human knowledge,​​ designed to be the​​​‌ final closed-ended academic benchmark‌ of its kind with‌​‌ broad subject coverage. HLE​​ consists of 3,000 questions​​​‌ across dozens of subjects,‌ including mathematics, humanities, and‌​‌ the natural sciences. HLE​​​‌ is developed globally by​ subject-matter experts and consists​‌ of multiple-choice and short-answer​​ questions suitable for automated​​​‌ grading. Each question has​ a known solution that​‌ is unambiguous and easily​​ verifiable, but cannot be​​​‌ quickly answered via internet​ retrieval. State-of-the-art LLMs demonstrate​‌ low accuracy and calibration​​ on HLE, highlighting a​​​‌ significant gap between current​ LLM capabilities and the​‌ expert human frontier on​​ closed-ended academic questions. To​​​‌ inform research and policymaking​ upon a clear understanding​‌ of model capabilities, we​​ publicly release HLE.​​​‌

Saturation-Driven Dataset Generation for​ LLM Mathematical Reasoning in​‌ the TPTP Ecosystem 50​​

The scarcity of high-quality,​​​‌ logically sound data is​ a critical bottleneck for​‌ advancing the mathematical reasoning​​ of Large Language Models​​​‌ (LLMs). Our work confronts​ this challenge by turning​‌ decades of automated theorem​​ proving research into a​​​‌ scalable data engine. Rather​ than relying on error-prone​‌ LLMs or complex proof-assistant​​ syntax like Lean and​​​‌ Isabelle, our framework leverages​ E-prover's saturation capabilities on​‌ the vast TPTP axiom​​ library to derive a​​​‌ massive, guaranteed-valid corpus of​ theorems. Our pipeline is​‌ principled and simple: saturate​​ axioms, filter for ”interesting”​​​‌ theorems, and generate tasks.​ With no LLMs in​‌ the loop, we eliminate​​ factual errors by construction.​​​‌ This purely symbolic data​ is then transformed into​‌ three difficulty-controlled challenges: entailment​​ verification, premise selection, and​​​‌ proof reconstruction. Our zero-shot​ experiments on frontier models​‌ reveal a clear weakness:​​ performance collapses on tasks​​​‌ requiring deep, structural reasoning.​ Our framework provides both​‌ the diagnostic tool to​​ measure this gap and​​​‌ a scalable source of​ symbolic training data to​‌ address it. We make​​ the code and data​​​‌ publicly available.

Logic​ Haystacks: Probing LLMs Long-Context​‌ Logical Reasoning (Without Easily​​ Identifiable Unrelated Padding) 53​​​‌

Large language models demonstrate​ promising long context processing​‌ capabilities, with recent models​​ touting context windows close​​​‌ to one million tokens.​ However, the evaluations supporting​‌ these claims often involve​​ simple retrieval tasks or​​​‌ synthetic tasks padded with​ irrelevant text, which the​‌ models may easily detect​​ and discard. In this​​​‌ work, we generate lengthy​ simplified English text with​‌ first-order logic representations spanning​​ up to 2048 clauses​​​‌ (around 25k GPT-4 tokens).​ We formulate an evaluation​‌ task with evidence retrieval​​ for contradiction detection. The​​​‌ long, homogeneous text is​ filled with distractors that​‌ are both hard to​​ distinguish from relevant evidences​​​‌ and provably not interfering​ with them. Our evaluation​‌ of evidence retrieval shows​​ that the effective context​​​‌ window is much smaller​ with realistic distractors, already​‌ crumbling at 128 clauses.​​

Generating Explanations in Medical​​​‌ Question-Answering by Expectation Maximization​ Inference over Evidence 24​‌

Medical Question Answering (medical​​ QA) systems play an​​​‌ essential role in assisting​ healthcare workers in finding​‌ answers to their questions.​​ However, it is not​​​‌ sufficient to merely provide​ answers by medical QA​‌ systems because users might​​ want explanations, that is,​​​‌ more analytic statements in​ natural language that describe​‌ the elements and context​​ that support the answer.​​​‌ To do so, we​ propose a novel approach​‌ for generating natural language​​ explanations for answers predicted​​ by medical QA systems.​​​‌ As high-quality medical explanations‌ require additional medical knowledge,‌​‌ so that our system​​ extracts knowledge from medical​​​‌ textbooks to enhance the‌ quality of explanations during‌​‌ the explanation generation process.​​ Concretely, we designed an​​​‌ Expectation-Maximization approach that makes‌ inferences about the evidence‌​‌ found in these texts,​​ offering an efficient way​​​‌ to focus attention on‌ lengthy evidence passages. Experimental‌​‌ results, conducted on two​​ datasets MQAE-diag and MQAE,​​​‌ demonstrate the effectiveness of‌ our framework for reasoning‌​‌ with textual evidence. Our​​ approach outperforms state-of-the-art models,​​​‌ achieving a significant improvement‌ of 6.13 and 5.47‌​‌ percentage points on the​​ Rouge-L score; 6.49 and​​​‌ 5.28 percentage points on‌ the Bleu-4 score on‌​‌ the MQAE-diag and MQAE​​ datasets.

Neural evidence for​​​‌ perceiving a vowel merger‌ after a social interaction‌​‌ within a native language​​ 17

Although previous research​​​‌ has shown that speakers‌ adapt on the words‌​‌ they use, it remains​​ unclear whether speakers adapt​​​‌ their phonological representations, leading‌ them to perceive new‌​‌ phonemic contrasts following a​​ social interaction. This event-related​​​‌ potential (ERP) study investigates‌ whether the neuronal responses‌​‌ to the perception of​​ the /e/-/ϵ/​​​‌ vowel merger in Northern‌ French speakers show evidence‌​‌ for discriminating /e/ and​​ /ϵ/ phonemes​​​‌ after interacting with a‌ speaker who produced this‌​‌ contrast. Northern French participants​​ engaged in an interactive​​​‌ map task and we‌ measured their ERP responses‌​‌ elicited after the presentation​​ of a last syllable​​​‌ which was either phonemically‌ identical to or different‌​‌ from preceding syllables. There​​ was no evidence for​​​‌ discrimination between /e/ and‌ /ϵ/ phonemes‌​‌ before the social interaction,​​ while mismatch negativity (MMN)​​​‌ and late responses revealed‌ /e/-/ϵ/ discrimination‌​‌ after the social interaction.​​ The findings suggest rapid​​​‌ neuronal adaptations of phonemic‌ representations thanks to the‌​‌ social interaction.

How does​​ the creation of new​​​‌ semantic relationships during dialogue‌ impact long-term semantic representations‌​‌ after dialogue? 19

Dialogue​​ is an ideal setting​​​‌ for changing linguistic representations‌ thanks to the repeated‌​‌ use of new words​​ and meanings. Two experiments​​​‌ were conducted to examine‌ the extent to which‌​‌ new semantic relationships created​​ during dialogue may change​​​‌ preexisting representations in long-term‌ semantic memory after a‌​‌ dialogue. For this purpose,​​ we developed an interactive​​​‌ agreement referential task to‌ create new semantic relationships‌​‌ in dialogue between two​​ words by associating them​​​‌ to a single picture.‌ One day after the‌​‌ dialogue phase, participants performed​​ either a lexical decision​​​‌ task associated to a‌ semantic priming paradigm (Experiment‌​‌ 1), or a semantic​​ relatedness judgment task (Experiment​​​‌ 2). In both tasks,‌ the participants' performance was‌​‌ collected during the processing​​ of pairs of words​​​‌ referring either to the‌ same picture or to‌​‌ different pictures during the​​ dialogue phase in order​​​‌ to assess changes in‌ long-term semantic representations after‌​‌ the dialogue phase. No​​ significant effect of relatedness​​​‌ due to the creation‌ of new semantic relationships‌​‌ during dialogue was found​​ in the lexical decision​​​‌ task. However, when the‌ participants' attention was focused‌​‌ on semantic relationships during​​​‌ the semantic relatedness judgment​ task, which required participants​‌ to perform an explicit​​ judgment, newly related words​​​‌ were rated as more​ related semantically. The two​‌ experiments bear important implications​​ for understanding on the​​​‌ links between dialogue and​ the updating of long-term​‌ semantic representations.

Indirect Reply​​ Processing in Multilingual Conversations​​​‌ when Inferring Speaker Meaning:​ an ERP study 37​‌, 56

For a​​ successful communication, interlocutors need​​​‌ to interpret an intended​ meaning beyond what is​‌ explicitly stated. Accordingly, a​​ pragmatic inference is often​​​‌ required, as is the​ case for indirect replies.​‌ For instance, if someone​​ replies to "What is​​​‌ it like giving a​ presentation?" with "Giving a​‌ good presentation is complicated",​​ the reply is direct.​​​‌ However, when responding to​ "Did you like my​‌ presentation?", the same reply​​ becomes indirect, implying the​​​‌ presentation was not well-received.​ An intriguing question is​‌ to know whether such​​ pragmatic inferences, required for​​​‌ indirect reply processing, are​ influenced by the cognitive​‌ demands of processing a​​ second language (L2).To explore​​​‌ this question, our experiment​ aimed to characterize the​‌ neural signature of indirect​​ reply processing in L1​​​‌ and L2 dialogues. We​ hypothesized that the additional​‌ cognitive cost of L2​​ processing would hinder pragmatic​​​‌ inferencing. To test this,​ we measured the event-related​‌ potentials (ERPs) of 40​​ French-speaking students listening to​​​‌ 144 dialogues in French​ (L1) and English (L2),​‌ in a within-participants design.​​ Each dialogue, preceded by​​​‌ a written context, ended​ with a direct or​‌ indirect reply. The ERPs​​ were time-locked on the​​​‌ final word of the​ reply. A visual inspection​‌ of the results revealed​​ that L1 dialogues elicited​​​‌ an early ERP effect,​ indicating faster semantic processing​‌ compared to L2 dialogues.​​ Initially, the response was​​​‌ more negative for direct​ replies; however, after 400​‌ ms, this negative response​​ became larger for indirect​​​‌ replies, reflecting an increased​ cognitive effort required for​‌ pragmatic inferences. In L2​​ dialogues, indirect replies elicited​​​‌ a persistent negative response​ after 400 ms, suggesting​‌ greater difficulty when inferring​​ the implied meaning in​​​‌ L2 dialogues.These findings highlight​ the additional cognitive demands​‌ of processing indirect replies,​​ particularly in L2 over​​​‌ late-processing stages. The current​ study therefore contributes to​‌ our understanding of pragmatic​​ processing and its neural​​​‌ underpinnings in multilingual communication.​

Interacting with someone shapes​‌ prediction in spoken-language comprehension​​ 54

Background: While listening​​​‌ to a spoken message,​ predicting upcoming information from​‌ a previous sentential context​​ ensures a successful comprehension​​​‌ [1,2]. However, prediction in​ spoken-language comprehension is not​‌ always found [3,4]. In​​ order to explain this​​​‌ flexibility, both prediction-by-association and​ prediction-by-production have been proposed​‌ [2]. While prediction-by-association would​​ be an automatic process​​​‌ related to the spreading​ activation of conceptual features​‌ from the sentence representation,​​ prediction-by-production, leading to the​​​‌ preactivation of words predicted​ from the sentence representation​‌ and their properties thanks​​ to the production system,​​​‌ would be optional. In​ this study, we investigated​‌ the flexibility of prediction-by-association​​ and prediction-by-production through a​​​‌ social interaction. This investigation​ is motivated by two​‌ key components of social​​ interactions: mutual comprehension at​​ the conceptual level and​​​‌ interplay between comprehension and‌ production systems.Method: In the‌​‌ Visual World Paradigm, thirty-two​​ native French speakers listened​​​‌ to forty-eight highly and‌ weakly constraining sentences which‌​‌ were associated with a​​ visual scene containing three​​​‌ distractor objects and one‌ of four critical objects‌​‌ (Target, Semantic Competitor, Phonological​​ Competitor, Unrelated). Each sentence​​​‌ was presented only once‌ before or after a‌​‌ social interaction, which was​​ not related to the​​​‌ content of the sentences‌ (see Figure 1A). The‌​‌ Target object referred to​​ the word predicted from​​​‌ a sentence whose effect‌ could reflect the two‌​‌ types of predictions. The​​ Semantic and Phonological Competitor​​​‌ objects referred respectively to‌ a word sharing either‌​‌ semantic category or phonological​​ onset overlap with the​​​‌ predicted word to explore‌ prediction-by-production. Used as a‌​‌ control condition, the Unrelated​​ object referred to a​​​‌ word with neither phonological‌ onset overlap nor semantic‌​‌ relationships with the predicted​​ word. Fixations on the​​​‌ different objects were recorded‌ during sentence comprehension. After‌​‌ hearing the sentence, participants​​ had to determine whether​​​‌ one of the objects‌ was mentioned in the‌​‌ sentence. During the social​​ interaction, participants had to​​​‌ find the correct position‌ for five Tangram pictures‌​‌ into a grid of​​ ten pictures under time​​​‌ pressure by actively collaborating‌ with their partner and‌​‌ describing shapes via an​​ audioconference device.Results: Cluster-based permutation​​​‌ analyses were performed on‌ fixation proportion differences between‌​‌ related and unrelated objects.​​ For highly constraining sentences,​​​‌ fixation proportions between Target‌ and Unrelated objects differed‌​‌ in a time window​​ between -320 and 980​​​‌ ms after word onset‌ before the interaction (p=.001,‌​‌ see Figure 1B). This​​ difference emerged 160 ms​​​‌ earlier after the interaction‌ in a time window‌​‌ between -480 ms and​​ 980 after word onset.​​​‌ Fixation proportions between Semantic‌ Competitor and Unrelated objects‌​‌ differed in a time​​ window between -380 and​​​‌ -80 ms after word‌ onset before the interaction‌​‌ (p=.004, see Figure 1B).​​ No other significant effects​​​‌ were found for highly‌ or weakly constraining sentences.Discussion:‌​‌ Consistent with optional prediction-by-production​​ [2], a predictive effect​​​‌ due to Semantic Competitor‌ objects only occurred before‌​‌ the social interaction. In​​ contrast, the predictive effect​​​‌ of Target objects was‌ speeded-up after the social‌​‌ interaction. This pilot study​​ suggests that social interaction​​​‌ may shift prediction toward‌ finer based-concept processing. Findings‌​‌ will be discussed in​​ line with models of​​​‌ social interaction and language‌ comprehension and further studies‌​‌ are needed to confirm​​ the role of interactions.​​​‌

Neural signature of indirect‌ reply processing while listening‌​‌ to foreign accented dialogues​​ 55

The interpretation of​​​‌ the sentence ‘Giving a‌ good presentation is complicated’‌​‌ will differ whether it​​ is a reply to​​​‌ (a) or (b):(a) What‌ is it like giving‌​‌ a presentation?(b) Did you​​ like my presentation?The neurocognitive​​​‌ mechanisms underlying the pragmatic‌ inference of indirect replies‌​‌ (e.g. in reply to​​ (b)), have mainly been​​​‌ studied when dialogues occur‌ in first language contexts.‌​‌ However, we hypothesize that​​ foreign-accented speech may affect​​​‌ such inferences given the‌ cognitive cost it generates‌​‌ (due to linguistic disfluency),​​​‌ as well as native​ listeners’ limited expectation of​‌ the foreign speaker’s linguistic​​ abilities. To test this​​​‌ hypothesis, our ongoing experiment​ aims to characterize the​‌ neural signature of indirect​​ reply processing in foreign-accented​​​‌ dialogues. Accordingly, we measure​ the event-related potentials (ERPs)​‌ of 40 French-speaking students​​ listening to native and​​​‌ foreign-accented dialogues. Each dialogue​ is preceded by a​‌ written context establishing the​​ communicative situation. A total​​​‌ of 144 dialogues are​ presented, ending in either​‌ a direct or indirect​​ reply (e.g. in reply​​​‌ to (a) or (b),​ respectively). For one-third of​‌ the dialogues, participants answer​​ yes-no comprehension questions.Based on​​​‌ the few previous ERP​ studies of indirect replies​‌ and foreign-accented pragmatics, we​​ expect to find ERPs​​​‌ time-locked to the final​ word of the reply​‌ that reflect additional cognitive​​ effort when processing foreign-accented​​​‌ indirect replies. Compared to​ native-accented direct replies, we​‌ anticipate a shallower processing.​​ Particularly, we should observe​​​‌ a smaller and delayed​ N400, indicating a hindered​‌ semantic integration of the​​ reply, and a larger​​​‌ P600, reflecting the additional​ pragmatic processing. Moreover, we​‌ will examine explanatory hypotheses​​ of how individual cognitive​​​‌ capacities, such as working​ memory and non-verbal reasoning,​‌ may modulate pragmatic processing​​ in multilingual communication, as​​​‌ previously observed in native-accented​ contexts. Our results will​‌ be discussed in line​​ with the neurocognitive models​​​‌ of language comprehension and​ pragmatics.

Faces and voices​‌ in dialogue: How partner-specific​​ cues contribute to conversational​​​‌ memory 38

Previous research​ suggests that information mentioned​‌ during dialogue is frequently​​ encoded in association with​​​‌ the current partner. This​ raises the question of​‌ which partner-specific cues might​​ contribute to the subsequent​​​‌ retrieval of information from​ memory. Following a joint​‌ communication task, individuals were​​ tested on recognition memory​​​‌ for referent labels in​ a context cued by​‌ their partner’s face and/or​​ voice. We examine whether​​​‌ partner-specific visual and auditory​ cues can facilitate access​‌ to information encoded during​​ conversation.

8.2 Privacy and​​​‌ NLP

Participants: Marc Tommasi​, Damien Sileo,​‌ Gabriel Loiseau.

Tau-Eval:​​ A Unified Evaluation Framework​​​‌ for Useful and Private​ Text Anonymization 32

Text​‌ anonymization is the process​​ of removing or obfuscating​​​‌ information from textual data​ to protect the privacy​‌ of individuals. This process​​ inherently involves a complex​​​‌ trade-off between privacy protection​ and information preservation, where​‌ stringent anonymization methods can​​ significantly impact the text's​​​‌ utility for downstream applications.​ Evaluating the effectiveness of​‌ text anonymization proves challenging​​ from both privacy and​​​‌ utility perspectives, as there​ is no universal benchmark​‌ that can comprehensively assess​​ anonymization techniques across diverse,​​​‌ and sometimes contradictory contexts.​ We present Tau-Eval, an​‌ open-source framework for benchmarking​​ text anonymization methods through​​​‌ the lens of privacy​ and utility task sensitivity.​‌ A Python library, code,​​ documentation and tutorials are​​​‌ publicly available.

TAROT: Task-Oriented​ Authorship Obfuscation Using Policy​‌ Optimization Methods 31

Authorship​​ obfuscation aims to disguise​​​‌ the identity of an​ author within a text​‌ by altering the writing​​ style, vocabulary, syntax, and​​​‌ other linguistic features associated​ with the text author.​‌ This alteration needs to​​ balance privacy and utility.​​ While strong obfuscation techniques​​​‌ can effectively hide the‌ author's identity, they often‌​‌ degrade the quality and​​ usefulness of the text​​​‌ for its intended purpose.‌ Conversely, maintaining high utility‌​‌ tends to provide insufficient​​ privacy, making it easier​​​‌ for an adversary to‌ de-anonymize the author. Thus,‌​‌ achieving an optimal trade-off​​ between these two conflicting​​​‌ objectives is crucial. In‌ this paper, we propose‌​‌ TAROT: Task-Oriented Authorship Obfuscation​​ Using Policy Optimization, a​​​‌ new unsupervised authorship obfuscation‌ method whose goal is‌​‌ to optimize the privacy-utility​​ trade-off by regenerating the​​​‌ entire text considering its‌ downstream utility. Our approach‌​‌ leverages policy optimization as​​ a fine-tuning paradigm over​​​‌ small language models in‌ order to rewrite texts‌​‌ by preserving author identity​​ and downstream task utility.​​​‌ We show that our‌ approach largely reduce the‌​‌ accuracy of attackers while​​ preserving utility. We make​​​‌ our code and models‌ publicly available.

8.3 Music‌​‌ and NLP

Participants: Mikaela​​ Keller, Dinh Viet-Toan​​​‌ Le.

Natural Language‌ Processing Methods for Symbolic‌​‌ Music Generation and Information​​ Retrieval: a Survey 21​​​‌

Several adaptations of Transformers‌ models have been developed‌​‌ in various domains since​​ its breakthrough in Natural​​​‌ Language Processing (NLP). This‌ trend has spread into‌​‌ the field of Music​​ Information Retrieval (MIR), including​​​‌ studies processing music data.‌ However, the practice of‌​‌ leveraging NLP tools for​​ symbolic music data is​​​‌ not novel in MIR.‌ Music has been frequently‌​‌ compared to language, as​​ they share several similarities,​​​‌ including sequential representations of‌ text and music. These‌​‌ analogies are also reflected​​ through similar tasks in​​​‌ MIR and NLP. This‌ survey reviews NLP methods‌​‌ applied to symbolic music​​ generation and information retrieval​​​‌ studies following two axes.‌ We first propose an‌​‌ overview of representations of​​ symbolic music adapted from​​​‌ natural language sequential representations.‌ Such representations are designed‌​‌ by considering the specificities​​ of symbolic music. These​​​‌ representations are then processed‌ by models. Such models,‌​‌ possibly originally developed for​​ text and adapted for​​​‌ symbolic music, are trained‌ on various tasks. We‌​‌ describe these models, in​​ particular deep learning models,​​​‌ through different prisms, highlighting‌ music-specialized mechanisms. We finally‌​‌ present a discussion surrounding​​ the effective use of​​​‌ NLP tools for symbolic‌ music data. This includes‌​‌ technical issues regarding NLP​​ methods and fundamental differences​​​‌ between text and music,‌ which may open several‌​‌ doors for further research​​ into more effectively adapting​​​‌ NLP tools to symbolic‌ MIR.

METEOR: Melody-aware Texture-controllable‌​‌ Symbolic Orchestral Music Generation​​ via Transformer VAE 30​​​‌

Re-orchestration is the process‌ of adapting a music‌​‌ piece for a different​​ set of instruments. By​​​‌ altering the original instrumentation,‌ the orchestrator often modifies‌​‌ the musical texture while​​ preserving a recognizable melodic​​​‌ line and ensures that‌ each part is playable‌​‌ within the technical and​​ expressive capabilities of the​​​‌ chosen instruments. In this‌ work, we propose METEOR,‌​‌ a model for generating​​ Melody-aware Texture-controllable re-Orchestration with​​​‌ a Transformer-based variational auto-encoder‌ (VAE). This model performs‌​‌ symbolic instrumental and textural​​ music style transfers with​​​‌ a focus on melodic‌ fidelity and controllability. We‌​‌ allow bar- and track-level​​​‌ controllability of the accompaniment​ with various textural attributes​‌ while keeping a homophonic​​ texture. With both subjective​​​‌ and objective evaluations, we​ show that our model​‌ outperforms style transfer models​​ on a re-orchestration task​​​‌ in terms of generation​ quality and controllability. Moreover,​‌ it can be adapted​​ for a lead sheet​​​‌ orchestration task as a​ zero-shot learning model, achieving​‌ performance comparable to a​​ model specifically trained for​​​‌ this task.

Evaluating Interval-based​ Tokenization for Pitch Representation​‌ in Symbolic Music Analysis​​ 57

Symbolic music analysis​​​‌ tasks are often performed​ by models originally developed​‌ for Natural Language Processing,​​ such as Transformers. Such​​​‌ models require the input​ data to be represented​‌ as sequences, which is​​ achieved through a process​​​‌ of tokenization. Tokenization strategies​ for symbolic music often​‌ rely on absolute MIDI​​ values to represent pitch​​​‌ information. However, music research​ largely promotes the benefit​‌ of higher-level representations such​​ as melodic contour and​​​‌ harmonic relations for which​ pitch intervals turn out​‌ to be more expressive​​ than absolute pitches. In​​​‌ this work, we introduce​ a general framework for​‌ building interval-based tokenizations. By​​ evaluating these tokenizations on​​​‌ three music analysis tasks,​ we show that such​‌ interval-based tokenizations improve model​​ performances and facilitate their​​​‌ explainability.

Modeling Symbolic Music​ with Natural Language Processing​‌ Approaches 41

Music is​​ often described as a​​​‌ language because of its​ similarities to natural language.​‌ These include their respective​​ representations through symbolic music​​​‌ notation and textual form.​ Therefore, the field of​‌ Music Information Retrieval (MIR)​​ has often borrowed several​​​‌ tools from the Natural​ Language Processing (NLP) field​‌ to adapt them to​​ process symbolic music data.​​​‌ In particular, this phenomenon​ has been increasingly popular​‌ with the breakthrough of​​ Transformer models in the​​​‌ NLP field.This thesis first​ provides a structured overview​‌ of adaptations of NLP​​ methods developed in the​​​‌ MIR field for symbolic​ music processing. They are​‌ presented along three axes,​​ each addressing the use​​​‌ of diverse representations of​ symbolic music at different​‌ levels. Symbolic music represented​​ as sequential data has​​​‌ lead to the development​ of several tokenization strategies,​‌ which we propose to​​ organize within a unified​​​‌ taxonomy. These representations are​ subsequently processed through models,​‌ such as recurrent or​​ attention-based architectures initially developed​​​‌ for text data, giving​ rise to multiple adaptations​‌ for symbolic music processing.​​ Finally, these abstract representations​​​‌ are used to perform​ tasks, where both parallels​‌ and distinctive characteristics emerge​​ between MIR and NLP.These​​​‌ aspects then structure the​ three technical contributions of​‌ this thesis. First, we​​ study the expressiveness of​​​‌ sequential representations of music​ through the development of​‌ interval-based tokenization strategies, and​​ the analysis of a​​​‌ subword tokenization strategy, Byte-Pair​ Encoding, applied to symbolic​‌ music tokens. We then​​ propose a framework for​​​‌ model explainability which leads​ to the analysis of​‌ the attention mechanism of​​ a Transformer-based model trained​​​‌ for functional harmony analysis.​ Finally, we develop a​‌ model adapted from NLP​​ tools for a task​​​‌ of re-orchestration, framed as​ a case of multi-track​‌ music generation.Ultimately, this thesis​​ defends that NLP methods​​ first remains a toolbox​​​‌ from which MIR studies‌ can take some tools‌​‌ from. Beyond the analogies​​ between music and natural​​​‌ language, the main motivation‌ guiding a MIR study‌​‌ should be musical questions.​​

8.4 Speech and Privacy​​​‌

Participants: Marc Tommasi.‌

Analysis of Speech Temporal‌​‌ Dynamics in the Context​​ of Speaker Verification and​​​‌ Voice Anonymization 35

In‌ this paper, we investigate‌​‌ the impact of speech​​ temporal dynamics in application​​​‌ to automatic speaker verification‌ and speaker voice anonymization‌​‌ tasks. We propose several​​ metrics to perform automatic​​​‌ speaker verification based only‌ on phoneme durations. Experimental‌​‌ results demonstrate that phoneme​​ durations leak some speaker​​​‌ information and can reveal‌ speaker identity from both‌​‌ original and anonymized speech.​​ Thus, this work emphasizes​​​‌ the importance of taking‌ into account the speaker's‌​‌ speech rate and, more​​ importantly, the speaker's phonetic​​​‌ duration characteristics, as well‌ as the need to‌​‌ modify them in order​​ to develop anonymization systems​​​‌ with strong privacy protection‌ capacity.

Exploiting Context-dependent Duration‌​‌ Features for Voice Anonymization​​ Attack Systems 36

The​​​‌ temporal dynamics of speech,‌ encompassing variations in rhythm,‌​‌ intonation, and speaking rate,​​ contain important and unique​​​‌ information about speaker identity.‌ This paper proposes a‌​‌ new method for representing​​ speaker characteristics by extracting​​​‌ context-dependent duration embeddings from‌ speech temporal dynamics. We‌​‌ develop novel attack models​​ using these representations and​​​‌ analyze the potential vulnerabilities‌ in speaker verification and‌​‌ voice anonymization systems.The experimental​​ results show that the​​​‌ developed attack models provide‌ a significant improvement in‌​‌ speaker verification performance for​​ both original and anonymized​​​‌ data in comparison with‌ simpler representations of speech‌​‌ temporal dynamics reported in​​ the literature.

8.5 Security​​​‌ and Privacy

Participants: Marc‌ Tommasi, Jan Ramon‌​‌, Antoine Barczewski,​​ Marc Damie, Clément​​​‌ Pierquin, Paul Andrey‌, Michaël Perrot,‌​‌ Jean Dufraiche.

Generalization​​ under Byzantine and Poisoning​​​‌ Attacks: Tight Stability Bounds‌ in Robust Distributed Learning‌​‌ 42

Robust distributed learning​​ algorithms aim to maintain​​​‌ good performance in distributed‌ and federated settings, even‌​‌ in the presence of​​ misbehaving workers. Two primary​​​‌ threat models have been‌ studied: Byzantine attacks, where‌​‌ misbehaving workers can send​​ arbitrarily corrupted updates, and​​​‌ data poisoning attacks, where‌ misbehavior is limited to‌​‌ manipulation of local training​​ data. While prior work​​​‌ has shown comparable optimization‌ error under both threat‌​‌ models, a fundamental question​​ remains open: How do​​​‌ these threat models impact‌ generalization? Empirical evidence suggests‌​‌ a gap between the​​ two threat models, yet​​​‌ it remains unclear whether‌ it is fundamental or‌​‌ merely an artifact of​​ suboptimal attacks. In this​​​‌ work, we present the‌ first theoretical investigation into‌​‌ this problem, formally showing​​ that Byzantine attacks are​​​‌ intrinsically more harmful to‌ generalization than data poisoning.‌​‌ Specifically, we prove that:​​ (i) under data poisoning,​​​‌ the uniform algorithmic stability‌ of a robust distributed‌​‌ learning algorithm, with optimal​​ optimization error, degrades by​​​‌ an additive factor of‌ (fn-‌​‌f), with​​ f the number of​​​‌ misbehaving workers out of‌ n; and (ii)‌​‌ In contrast, under Byzantine​​​‌ attacks, the degradation is​ in 𝒪fn​‌-2f.This​​ difference in stability leads​​​‌ to a generalization error​ gap that is especially​‌ significant as f approaches​​ its maximum value n​​​‌2.

Privacy-Preserving Computations​ on Sparse Data 40​‌

Data breaches and privacy​​ violations have raised global​​​‌ concerns about the protection​ of personal information. To​‌ address these concerns, cryptographic​​ protocols known as Multi-Party​​​‌ Computations (MPC) have been​ developed to enable multiple​‌ parties to jointly compute​​ on their private inputs​​​‌ without revealing them. These​ protocols have found applications​‌ notably in tax fraud​​ detection, healthcare, and machine​​​‌ learning. However, existing MPC​ protocols remain insufficient for​‌ many real-world scenarios, particularly​​ when dealing with high-dimensional​​​‌ or structured data.This thesis​ focuses on sparse data;​‌ datasets containing mostly zero​​ values, which naturally arise​​​‌ in applications such as​ recommender systems and healthcare.​‌ While plaintext algorithms for​​ sparse data are well​​​‌ established, few cryptographic protocols​ are optimized for this​‌ setting, limiting the practicality​​ of MPC in domains​​​‌ where sparsity is the​ norm.The core contributions of​‌ this thesis introduce new​​ cryptographic protocols tailored to​​​‌ sparse computations. We propose​ MPC protocols for secure​‌ sparse matrix multiplication, enabling​​ performance gains that make​​​‌ previously impractical applications feasible.​ We also design new​‌ Function Secret Sharing (FSS)​​ schemes able to efficiently​​​‌ aggregate sparse data.Beyond these​ protocols, the thesis makes​‌ several orthogonal contributions that​​ question key assumptions and​​​‌ practical aspects of privacy-enhancing​ technologies. We study the​‌ role of leakage in​​ sparse computations by analyzing​​​‌ access-pattern leakage in searchable​ encryption, providing new attacks​‌ and statistical insights into​​ its risks. We present​​​‌ Fedivertex, a new graph​ dataset based on decentralized​‌ social media, to benchmark​​ decentralized machine learning. Finally,​​​‌ we evaluate the energy​ consumption of several privacy-enhancing​‌ technologies.Taken together, these contributions​​ both advance the design​​​‌ of cryptographic protocols for​ sparse data and provide​‌ a broader perspective on​​ the challenges of deploying​​​‌ privacy-preserving computations in practice.​

Noisy Function Secret Sharing​‌ and its applications to​​ Differentially Private computations 45​​​‌

Function Secret Sharing (FSS)​ schemes enable to share​‌ secret functions between multiple​​ parties, with notable applications​​​‌ in anonymous communication and​ privacy-preserving machine learning. While​‌ two-party schemes offer logarithmic​​ key sizes, multi-party schemes​​​‌ remain less practical due​ to significantly larger keys.​‌ Although several approaches have​​ been proposed to improve​​​‌ multi-party schemes, a significant​ efficiency gap remains between​‌ the two-party and multi-party​​ settings.

Our work introduces​​​‌ noisy FSS: a relaxation​ of FSS preserving the​‌ standard privacy guarantees but​​ relaxing the correctness definition​​​‌ by allowing a small​ amount of noise in​‌ the output. We formally​​ define noisy FSS and​​​‌ show how the noise​ introduced by the scheme​‌ can be leveraged to​​ provide differential private outputs​​​‌ in statistics applications.

To​ demonstrate the benefits of​‌ this relaxation, we adapt​​ a scheme proposed by​​​‌ Corrigan-Gibbs et al. (S&P'15).​ While their scheme provides​‌ the smallest key sizes​​ among multi-party schemes, they​​​‌ do not support some​ applications notably in statistics​‌ due to their non-linear​​ share decoding. On the​​ contrary, recent works such​​​‌ as Goel et al.‌ (CRYPTO'25) have larger keys,‌​‌ but support all FSS​​ applications. Our noisy adapted​​​‌ scheme offers the best‌ of both worlds by‌​‌ matching the best key​​ sizes, while providing the​​​‌ properties necessary to statistics‌ applications.

How to Securely‌​‌ Shuffle? A survey about​​ Secure Shufflers for privacy-preserving​​​‌ computations 43

Ishai et‌ al. (FOCS'06) introduced secure‌​‌ shuffling as an efficient​​ building block for private​​​‌ data aggregation. Recently, the‌ field of differential privacy‌​‌ has revived interest in​​ secure shufflers by highlighting​​​‌ the privacy amplification they‌ can provide in various‌​‌ computations. Although several works​​ argue for the utility​​​‌ of secure shufflers, they‌ often treat them as‌​‌ black boxes; overlooking the​​ practical vulnerabilities and performance​​​‌ trade-offs of existing implementations.‌ This leaves a central‌​‌ question open: what makes​​ a good secure shuffler?​​​‌ This survey addresses that‌ question by identifying, categorizing,‌​‌ and comparing 26 secure​​ protocols that realize the​​​‌ necessary shuffling functionality. To‌ enable a meaningful comparison,‌​‌ we adapt and unify​​ existing security definitions into​​​‌ a consistent set of‌ properties. We also present‌​‌ an overview of privacy-preserving​​ technologies that rely on​​​‌ secure shufflers, offer practical‌ guidelines for selecting appropriate‌​‌ protocols, and outline promising​​ directions for future work.​​​‌

Revisiting the Attacker’s Knowledge‌ in Inference Attacks Against‌​‌ Searchable Symmetric Encryption 27​​

Encrypted search schemes have​​​‌ been proposed to address‌ growing privacy concerns. However,‌​‌ several leakage-abuse attacks have​​ highlighted some security vulnerabilities.​​​‌ Recent attacks assumed an‌ attacker's knowledge containing data‌​‌ “similar” to the indexed​​ data. However, this vague​​​‌ assumption is barely discussed‌ in literature: how likely‌​‌ is it for an​​ attacker to obtain a​​​‌ "similar enough" data? Our‌ paper provides novel statistical‌​‌ tools usable on any​​ attack in this setting​​​‌ to analyze its sensitivity‌ to data similarity. First,‌​‌ we introduce a mathematical​​ model based on statistical​​​‌ estimators to analytically understand‌ the attackers' knowledge and‌​‌ the notion of similarity.​​ Second, we conceive statistical​​​‌ tools to model the‌ influence of the similarity‌​‌ on the attack accuracy.​​ We apply our tools​​​‌ on three existing attacks‌ to answer questions such‌​‌ as: is similarity the​​ only factor influencing accuracy​​​‌ of a given attack?‌ Third, we show that‌​‌ the enforcement of a​​ maximum index size can​​​‌ make the “similar-data” assumption‌ harder to satisfy. In‌​‌ particular, we propose a​​ statistical method to estimate​​​‌ an appropriate maximum size‌ for a given attack‌​‌ and dataset. For the​​ best known attack on​​​‌ the Enron dataset, a‌ maximum index size of‌​‌ 200 guarantees (with high​​ probability) the attack accuracy​​​‌ to be below 5%.‌

Secure Sparse Matrix Multiplications‌​‌ and their Applications to​​ Privacy-Preserving Machine Learning 44​​​‌

To preserve privacy, multi-party‌ computation (MPC) enables executing‌​‌ Machine Learning (ML) algorithms​​ on secret-shared or encrypted​​​‌ data. However, existing MPC‌ frameworks are not optimized‌​‌ for sparse data. This​​ makes them unsuitable for​​​‌ ML applications involving sparse‌ data, e.g., recommender systems‌​‌ or genomics. Even in​​ plaintext, such applications involve​​​‌ high-dimensional sparse data, that‌ cannot be processed without‌​‌ sparsity-related optimizations due to​​​‌ prohibitively large memory requirements.​ Since matrix multiplication is​‌ central in ML algorithms,​​ we propose MPC algorithms​​​‌ to multiply secret sparse​ matrices. On the one​‌ hand, our algorithms avoid​​ the memory issues of​​​‌ the ”dense” data representation​ of classic secure matrix​‌ multiplication algorithms. On the​​ other hand, our algorithms​​​‌ can significantly reduce communication​ costs (some experiments show​‌ a factor 1000) for​​ realistic problem sizes. We​​​‌ validate our algorithms in​ two ML applications in​‌ which existing protocols are​​ impractical. An important question​​​‌ when developing MPC algorithms​ is what assumptions can​‌ be made. In our​​ case, if the number​​​‌ of non-zeros in a​ row is a sensitive​‌ piece of information then​​ a short runtime may​​​‌ reveal that the number​ of non-zeros is small.​‌ Existing approaches make relatively​​ simple assumptions, e.g., that​​​‌ there is a universal​ upper bound to the​‌ number of non-zeros in​​ a row. This often​​​‌ doesn't align with statistical​ reality, in a lot​‌ of sparse datasets the​​ amount of data per​​​‌ instance satisfies a power​ law. We propose an​‌ approach which allows adopting​​ a safe upper bound​​​‌ on the distribution of​ non-zeros in rows/columns of​‌ sparse matrices.

Dropout-Robust Mechanisms​​ for Differentially Private and​​​‌ Fully Decentralized Mean Estimation​ 52

Achieving differentially private​‌ computations in decentralized settings​​ poses significant challenges, particularly​​​‌ regarding accuracy, communication cost,​ and robustness against information​‌ leakage. While cryptographic solutions​​ offer promise, they often​​​‌ suffer from high communication​ overhead or require centralization​‌ in the presence of​​ network failures. Conversely, existing​​​‌ fully decentralized approaches typically​ rely on relaxed adversarial​‌ models or pairwise noise​​ cancellation, the latter suffering​​​‌ from substantial accuracy degradation​ if parties unexpectedly disconnect.​‌ In this work, we​​ propose IncA, a new​​​‌ protocol for fully decentralized​ mean estimation, a widely​‌ used primitive in data-intensive​​ processing. Our protocol, which​​​‌ enforces differential privacy, requires​ no central orchestration and​‌ employs low-variance correlated noise,​​ achieved by incrementally injecting​​​‌ sensitive information into the​ computation. First, we theoretically​‌ demonstrate that, when no​​ parties permanently disconnect, our​​​‌ protocol achieves accuracy comparable​ to that of a​‌ centralized setting-already an improvement​​ over most existing decentralized​​​‌ differentially private techniques. Second,​ we empirically show that​‌ our use of low-variance​​ correlated noise significantly mitigates​​​‌ the accuracy loss experienced​ by existing techniques in​‌ the presence of dropouts.​​

TAMIS: Tailored Membership Inference​​​‌ Attacks on Synthetic Data​ 25

<div><p>Membership Inference Attacks​‌ (MIA) enable to empirically​​ assess the privacy of​​​‌ a machine learning algorithm.​ In this paper, we​‌ propose TAMIS, a novel​​ MIA against differentially-private synthetic​​​‌ data generation methods that​ rely on graphical models.​‌ This attack builds upon​​ MAMA-MIA, a recently-published state-of-the-art​​​‌ method. It lowers its​ computational cost and requires​‌ less attacker knowledge. Our​​ attack is the product​​​‌ of a two-fold improvement.​ First, we recover the​‌ graphical model having generated​​ a synthetic dataset by​​​‌ using solely that dataset,​ rather than shadow-modeling over​‌ an auxiliary one. This​​ proves less costly and​​​‌ more performant. Second, we​ introduce a more mathematically-grounded​‌ attack score, that provides​​ a natural threshold for​​ binary predictions. In our​​​‌ experiments, TAMIS achieves better‌ or similar performance as‌​‌ MAMA-MIA on replicas of​​ the SNAKE challenge.

Privacy​​​‌ Amplification Through Synthetic Data:‌ Insights from Linear Regression‌​‌ 34

Synthetic data inherits​​ the differential privacy guarantees​​​‌ of the model used‌ to generate it. Additionally,‌​‌ synthetic data may benefit​​ from privacy amplification when​​​‌ the generative model is‌ kept hidden. While empirical‌​‌ studies suggest this phenomenon,​​ a rigorous theoretical understanding​​​‌ is still lacking. In‌ this paper, we investigate‌​‌ this question through the​​ well-understood framework of linear​​​‌ regression. First, we establish‌ negative results showing that‌​‌ if an adversary controls​​ the seed of the​​​‌ generative model, a single‌ synthetic data point can‌​‌ leak as much information​​ as releasing the model​​​‌ itself. Conversely, we show‌ that when synthetic data‌​‌ is generated from random​​ inputs, releasing a limited​​​‌ number of synthetic data‌ points amplifies privacy beyond‌​‌ the model's inherent guarantees.​​ We believe our findings​​​‌ in linear regression can‌ serve as a foundation‌​‌ for deriving more general​​ bounds in the future.​​​‌

Enhancing Differentially private machine‌ learning: Optimizations for Repeated‌​‌ Query scenarios 39

Deep​​ neural networks and other​​​‌ machine learning models have‌ experienced unprecedented growth in‌​‌ recent years. Alongside this​​ enthusiasm, there has been​​​‌ an increasing and well-founded‌ concern about the privacy‌​‌ of the vast amounts​​ of data required to​​​‌ train these models. The‌ combination of these two‌​‌ factors has been a​​ key driver of interest​​​‌ in privacy-preserving machine learning‌ techniques. Differential Privacy has‌​‌ emerged as the gold​​ standard for measuring privacy.​​​‌ This framework is now‌ applied on a wide‌​‌ range of data-driven tasks,​​ such as machine learning​​​‌ and collaborative analysis, where‌ multiple stakeholders wish to‌​‌ query shared data without​​ exposing their own. The​​​‌ main challenge in this‌ domain lies in balancing‌​‌ privacy guarantees with the​​ utility of the results.​​​‌ Indeed, privacy-preserving techniques often‌ come at the cost‌​‌ of reduced utility.This thesis​​ focuses on techniques to​​​‌ improve machine learning models‌ and tools foranalyzing them,‌​‌ while ensuring a satisfactory​​ level of privacy for​​​‌ the underlying data.First, it‌ introduces an innovative approach‌​‌ to privacy-preserving gradient descentmethods​​ by addressing the bias​​​‌ introduced by existing methods.‌ By leveraging prop-erties of‌​‌ gradient regularity rather than​​ clipping the gradient, as​​​‌ it is commonly donein‌ popular methods, our approach‌​‌ effectively reduces bias and​​ the noise added tothe​​​‌ gradient. We propose a‌ new algorithm that surpasses‌​‌ the state of the​​ art acrossvarious datasets.Second, the​​​‌ thesis explores techniques for‌ computing privacy-preserving empiricalcummulative distribution‌​‌ functions, even in cases​​ where the data is​​​‌ distributed acrossmultiple entities. This‌ study proposes a novel‌​‌ method compatible with different​​ secu-rity protocols, offering provable​​​‌ privacy guarantees and an‌ analysis of computationalcosts. range‌​‌ of applications are explored,​​ and experimental results are​​​‌ presented tovalidate the utility‌ of these methods.By analyzing‌​‌ optimization mechanisms and distribution​​ functions, this thesis con-tributes​​​‌ to the development of‌ more practical and efficient‌​‌ privacy-preserving machinelearning and data​​ analysis techniques.

Evaluating Membership​​​‌ Inference Attacks in Heterogeneous-Data‌ Setups 28

Among all‌​‌ privacy attacks against Machine​​​‌ Learning (ML), membership inference​ attacks (MIA) attracted the​‌ most attention. In these​​ attacks, the attacker is​​​‌ given an ML model​ and a data point,​‌ and they must infer​​ whether the data point​​​‌ was used for training.​ The attacker also has​‌ an auxiliary dataset to​​ tune their inference algorithm.​​​‌ Attack papers commonly simulate​ setups in which the​‌ attacker's and the target's​​ datasets are sampled from​​​‌ the same distribution. This​ setting is convenient to​‌ perform experiments, but it​​ rarely holds in practice.​​​‌ ML literature commonly starts​ with similar simplifying assumptions​‌ (i.e., "i.i.d." datasets), and​​ later generalizes the results​​​‌ to support heterogeneous data​ distributions. Similarly, our work​‌ makes a first step​​ in the generalization of​​​‌ the MIA evaluation to​ heterogeneous data. First, we​‌ design a metric to​​ measure the heterogeneity between​​​‌ any pair of tabular​ data distributions. This metric​‌ provides a continuous scale​​ to analyze the phenomenon.​​​‌ Second, we compare two​ methodologies to simulate a​‌ data heterogeneity between the​​ target and the attacker.​​​‌ These setups provide opposite​ performances: 90% attack accuracy​‌ vs. 50% (i.e., random​​ guessing). Our results show​​​‌ that the MIA accuracy​ depends on the experimental​‌ setup; and even if​​ research on MIA considers​​​‌ heterogeneous data setups, we​ have no standardized baseline​‌ of how to simulate​​ it. The lack of​​​‌ such a baseline for​ MIA experiments poses a​‌ significant challenge to risk​​ assessments in real-world machine​​​‌ learning scenarios.

Learning with​ Locally Private Examples by​‌ Inverse Weierstrass Private Stochastic​​ Gradient Descent (in preparation)​​​‌

Local Differential Privacy (LDP)​ has emerged as a​‌ prevailing framework for publishing​​ data privately without relying​​​‌ on a central trusted​ authority. In this paper,​‌ we are interested in​​ such a setting where​​​‌ examples are first privatized​ using the Gaussian and​‌ Randomized Response mechanisms and​​ then publicly released. We​​​‌ first leverage tools from​ the Gaussian smoothing literature,​‌ namely the Weierstrass transform,​​ to characterize the bias​​​‌ that standard risk minimization​ on such privatized data​‌ would induce. To mitigate​​ this bias, we then​​​‌ propose the Inverse Weierstrass​ Private SGD algorithm (IWP-SGD),​‌ a variant of stochastic​​ gradient descent that is​​​‌ guaranteed to recover, in​ expectation, the solution to​‌ the original, non-biased problem.​​ This new method leverages​​​‌ the invertibility of the​ Weierstrass transform to build​‌ an unbiased estimator of​​ the gradient with finite​​​‌ variance. It allows us​ to derive non-asymptotic convergence​‌ rates. Empirically, we validate​​ it on binary problems​​​‌ on synthetic and real​ data.

8.6 Fairness

Participants:​‌ Michaël Perrot, Marc​​ Tommasi, Shreya Venugopal​​​‌.

Fair Text Classification​ via Transferable Representations 22​‌

Group fairness is a​​ central research topic in​​​‌ text classification, where reaching​ fair treatment between sensitive​‌ groups (e.g., women and​​ men) remains an open​​​‌ challenge. We propose an​ approach that extends the​‌ use of the Wasserstein​​ Dependency Measure for learning​​​‌ unbiased neural text classifiers.​ Given the challenge of​‌ distinguishing fair from unfair​​ information in a text​​​‌ encoder, we draw inspiration​ from adversarial training by​‌ inducing independence between representations​​ learned for the target​​ label and those for​​​‌ a sensitive attribute. We‌ further show that Domain‌​‌ Adaptation can be efficiently​​ leveraged to remove the​​​‌ need for access to‌ the sensitive attributes in‌​‌ the dataset we cure.​​ We provide both theoretical​​​‌ and empirical evidence that‌ our approach is well-founded.‌​‌

Preserving Fairness when Making​​ Stochastic Predictions Deterministic (in​​​‌ preparation)

Deterministic decisions are‌ desirable in many high-stake‌​‌ automated decision processes. The​​ usual solution to achieve​​​‌ that when only stochastic‌ classifiers are available is‌​‌ to use 0.5 thresholding.​​ Unfortunately, this process may​​​‌ induce additional unfairness. In‌ this paper, we propose‌​‌ the first method specifically​​ tailored to make stochastic​​​‌ predictions deterministic while preserving‌ the fairness level of‌​‌ the stochastic classifier, that​​ is without introducing any​​​‌ additional bias. Leveraging ideas‌ from the post-processing literature,‌​‌ we demonstrate that our​​ method is theoretically sound.​​​‌ It comes with generalization‌ guarantees for both fairness‌​‌ and accuracy. Furthermore, it​​ inherits asymptotic optimality properties.​​​‌

Empirically, we show that‌ our method preserves fairness‌​‌ well for several base​​ stochastic classifiers and datasets.​​​‌ We also show that‌ it constitutes a new‌​‌ competitive avenue to learn​​ fair deterministic decision models.​​​‌

8.7 Federated and Decentralized‌ Learning

Participants: Batiste Le‌​‌ Bars, Marc Damie​​, Aleksei Korneev,​​​‌ Marc Tommasi.

Fedivertex:‌ a Graph Dataset based‌​‌ on Decentralized Social Networks​​ for Trustworthy Machine Learning​​​‌ 26

Decentralized machine learning‌ - where each client‌​‌ keeps its own data​​ locally and uses its​​​‌ own computational resources to‌ collaboratively train a model‌​‌ by exchanging peer-to-peer messages​​ - is increasingly popular,​​​‌ as it enables better‌ scalability and control over‌​‌ the data. A major​​ challenge in this setting​​​‌ is that learning dynamics‌ depend on the topology‌​‌ of the communication graph,​​ which motivates the use​​​‌ of real graph datasets‌ for benchmarking decentralized algorithms.‌​‌ Unfortunately, existing graph datasets​​ are largely limited to​​​‌ for-profit social networks crawled‌ at a fixed point‌​‌ in time and often​​ collected at the user​​​‌ scale, where links are‌ heavily influenced by the‌​‌ platform and its recommendation​​ algorithms. The Fediverse, which​​​‌ includes several free and‌ open-source decentralized social media‌​‌ platforms such as Mastodon,​​ Misskey, and Lemmy, offers​​​‌ an interesting real-world alternative.‌ We introduce Fedivertex, a‌​‌ new dataset of 182​​ graphs, covering seven social​​​‌ networks from the Fediverse,‌ crawled weekly over 14‌​‌ weeks. We release the​​ dataset along with a​​​‌ Python package to facilitate‌ its use, and illustrate‌​‌ its utility on several​​ tasks, including a new​​​‌ defederation task, which captures‌ a process of link‌​‌ deletion observed on these​​ networks.

A Survey on​​​‌ Verifiable Cross-Silo Federated Learning‌ 20

Federated Learning (FL)‌​‌ is a widespread approach​​ that allows training machine​​​‌ learning (ML) models with‌ data distributed across multiple‌​‌ storage units. In cross-silo​​ FL, which often appears​​​‌ in domains like healthcare‌ or finance, the number‌​‌ of participants is moderate,​​ and each party typically​​​‌ represents a well-known organization.‌ For instance, in medicine‌​‌ data owners are often​​ hospitals or data hubs​​​‌ which are well-established entities.‌ However, malicious parties may‌​‌ still attempt to disturb​​​‌ the training procedure in​ order to obtain certain​‌ benefits, for example, a​​ biased result or a​​​‌ reduction in computational load.​ While one can easily​‌ detect a malicious agent​​ when data used for​​​‌ training is public, the​ problem becomes much more​‌ acute when it is​​ necessary to maintain the​​​‌ privacy of the training​ dataset. To address this​‌ issue, there is recently​​ growing interest in developing​​​‌ verifiable protocols, where one​ can check that parties​‌ do not deviate from​​ the training procedure and​​​‌ perform computations correctly. In​ this paper, we present​‌ a survey on verifiable​​ cross-silo FL. We analyze​​​‌ various protocols, fit them​ in a taxonomy, and​‌ compare their efficiency and​​ threat models. We also​​​‌ analyze Zero-Knowledge Proof (ZKP)​ schemes and discuss how​‌ their overall cost in​​ a FL context can​​​‌ be minimized. Lastly, we​ identify research gaps and​‌ discuss potential directions for​​ future scientific work.

Adaptive​​​‌ collaboration for online personalized​ distributed learning with heterogeneous​‌ clients 49

We study​​ the problem of online​​​‌ personalized decentralized learning with​ N statistically heterogeneous clients​‌ collaborating to accelerate local​​ training. An important challenge​​​‌ in this setting is​ to select relevant collaborators​‌ to reduce gradient variance​​ while mitigating the introduced​​​‌ bias. To tackle this,​ we introduce a gradient-based​‌ collaboration criterion, allowing each​​ client to dynamically select​​​‌ peers with similar gradients​ during the optimization process.​‌ Our criterion is motivated​​ by a refined and​​​‌ more general theoretical analysis​ of the All-for-one algorithm,​‌ proved to be optimal​​ in Even et al.​​​‌ (2022) for an oracle​ collaboration scheme. We derive​‌ excess loss upper-bounds for​​ smooth objective functions, being​​​‌ either strongly convex, non-convex,​ or satisfying the Polyak-Łojasiewicz​‌ condition; our analysis reveals​​ that the algorithm acts​​​‌ as a variance reduction​ method where the speed-up​‌ depends on a sufficient​​ variance. We put​​​‌ forward two collaboration methods​ instantiating the proposed general​‌ schema; and we show​​ that one variant preserves​​​‌ the optimality of All-for-one​. We validate our​‌ results with experiments on​​ synthetic and real datasets.​​​‌

Federated Learning for MRI-based​ BrainAGE: a multicenter study​‌ on post-stroke functional outcome​​ prediction 51

Objective: Brain-predicted​​​‌ age difference (BrainAGE) is​ a neuroimaging biomarker reflecting​‌ brain health. However, training​​ robust BrainAGE models requires​​​‌ large datasets, often restricted​ by privacy concerns. This​‌ study evaluates the performance​​ of federated learning (FL)​​​‌ for BrainAGE estimation in​ ischemic stroke patients treated​‌ with mechanical thrombectomy, and​​ investigates its association with​​​‌ clinical phenotypes and functional​ outcomes.Methods: We used FLAIR​‌ brain images from 1674​​ stroke patients across 16​​​‌ hospital centers. We implemented​ standard machine learning and​‌ deep learning models for​​ BrainAGE estimates under three​​​‌ data management strategies: centralized​ learning (pooled data), FL​‌ (local training at each​​ site), and single-site learning.​​​‌ We reported prediction errors​ and examined associations between​‌ BrainAGE and vascular risk​​ factors (e.g., diabetes mellitus,​​​‌ hypertension, smoking), as well​ as functional outcomes at​‌ three months post-stroke. Logistic​​ regression evaluated BrainAGE's predictive​​​‌ value for these outcomes,​ adjusting for age, sex,​‌ vascular risk factors, stroke​​ severity, time between MRI​​ and arterial puncture, prior​​​‌ intravenous thrombolysis, and recanalisation‌ outcome.Results: While centralized learning‌​‌ yielded the most accurate​​ predictions, FL consistently outperformed​​​‌ single-site models. BrainAGE was‌ significantly higher in patients‌​‌ with diabetes mellitus across​​ all models. Comparisons between​​​‌ patients with good and‌ poor functional outcomes, and‌​‌ multivariate predictions of these​​ outcomes showed the significance​​​‌ of the association between‌ BrainAGE and post-stroke recovery.Conclusion:‌​‌ FL enables accurate age​​ predictions without data centralization.​​​‌ The strong association between‌ BrainAGE, vascular risk factors,‌​‌ and post-stroke recovery highlights​​ its potential for prognostic​​​‌ modeling in stroke care.‌

8.8 Conformal Prediction for‌​‌ Trustworthy Machine Learning

Participants:​​ Batiste Le Bars.​​​‌

On Volume Minimization in‌ Conformal Regression 29

We‌​‌ study the question of​​ volume optimality in split​​​‌ conformal regression, a topic‌ still poorly understood in‌​‌ comparison to coverage control.​​ Using the fact that​​​‌ the calibration step can‌ be seen as an‌​‌ empirical volume minimization problem,​​ we first derive a​​​‌ finite-sample upper-bound on the‌ excess volume loss of‌​‌ the interval returned by​​ the classical split method.​​​‌ This important quantity measures‌ the difference in length‌​‌ between the interval obtained​​ with the split method​​​‌ and the shortest oracle‌ prediction interval. Then, we‌​‌ introduce EffOrt, a methodology​​ that modifies the learning​​​‌ step so that the‌ base prediction function is‌​‌ selected in order to​​ minimize the length of​​​‌ the returned intervals. In‌ particular, our theoretical analysis‌​‌ of the excess volume​​ loss of the prediction​​​‌ sets produced by EffOrt‌ reveals the links between‌​‌ the learning and calibration​​ steps, and notably the​​​‌ impact of the choice‌ of the function class‌​‌ of the base predictor.​​ We also introduce Ad-EffOrt,​​​‌ an extension of the‌ previous method, which produces‌​‌ intervals whose size adapts​​ to the value of​​​‌ the covariate. Finally, we‌ evaluate the empirical performance‌​‌ and the robustness of​​ our methodologies.

9 Bilateral​​​‌ contracts and grants with‌ industry

9.1 Bilateral contracts‌​‌ with industry

We have​​ started two new CIFRE​​​‌ contracts in 2023. We‌ continue these collaborations until‌​‌ 2026.

Transfer learning for​​ text anonymization

Participants: Damien​​​‌ Sileo, Marc Tommasi‌, Gabriel Loiseau.‌​‌

VADE is a major​​ company that processes emails​​​‌ at large scale to‌ detect attacks like phishing.‌​‌

In this project we​​ design utility and privacy​​​‌ evaluation methods based on‌ the combination of many‌​‌ tasks and objectives, relevant​​ in the text (email)​​​‌ context. We study and‌ compare approaches based on‌​‌ text generation or based​​ on the replacement or​​​‌ obfuscation of selected entities,‌ to tune the privacy‌​‌ utility trade-off.

Synthetic data​​ generation with privacy constraints​​​‌

Participants: Aurélien Bellet,‌ Marc Tommasi, Clément‌​‌ Pierquin.

Craft.ai is​​ a company whose activity​​​‌ was originally focused on‌ explainable models for time‌​‌ series. It offers now​​ MLops solutions based on​​​‌ AI with trustworthy guarantees.‌ In this bilateral project‌​‌ with Craft.ai, Magnet brings​​ expertise in privacy preserving​​​‌ machine learning for the‌ generation of synthetic data.‌​‌

The project is organized​​ in four major axes.​​​‌ The definition of quality‌ metrics for synthetic data;‌​‌ the design of algorithms​​​‌ for synthetic data generation​ with differential privacy guarantees;​‌ the definition of theoretical​​ and empirical bounds on​​​‌ privacy associated with the​ release of synthetic data​‌ sets or generative models;​​ some applications on time​​​‌ series or correlated data.​

10 Partnerships and cooperations​‌

10.1 International research visitors​​

10.1.1 Visits of international​​​‌ scientists

Inria International Chair​ Nicolas Papernot : AI​‌ Treaties
  • Host teams: PreMeDICaL​​ (primary host), Magnet, Privatics.​​​‌

Participants: Nicolas Papernot,​ Marc Tommasi, Michaël​‌ Perrot, Raouf Kerkouche​​, Jan Ramon,​​​‌ Damien Sileo, Pascal​ Denis, Mikaela Keller​‌, Batiste Le Bars​​.

The research agenda​​​‌ during the term of​ the Inria Chair will​‌ thus be structured around​​ three major research thrusts,​​​‌ which will inform each​ other:

  1. Towards differentially private​‌ decentralized learning with applications​​ to healthcare.
  2. How can​​​‌ learning algorithms be co-designed​ with cryptographic protocols to​‌ obtain verifiable certificates at​​ the conclusion of training?​​​‌
  3. How will competition, cooperation,​ and coordination between countries,​‌ companies impact the stability​​ of AI governance?

10.1.2​​​‌ Visits to international teams​

Research stays abroad

Participant:​‌ Dinh Viet-Toan Le.​​

10.2 European initiatives

10.2.1​ Horizon Europe

TRUMPET

Participant:​‌ Jan Ramon [contact person]​​.

  • Title:
    TRUstworthy Multi-site​​​‌ Privacy Enhancing Technologies
  • Duration:​
    From October 1, 2022​‌ to December 31, 2025​​
  • Partners:
    • INRIA, France
    • TIMELEX,​​​‌ Belgium
    • Technovative Solutions LTD,​ United Kingdom
    • Fundacion Centro​‌ Tecnoloxico de Telecomunicacions de​​ Galicia (GRADIANT), Spain
    • Commissariat​​​‌ à l'Énergie Atomique et​ aux Énergies alternatives (CEA),​‌ France
    • Istituto Romagnolo per​​ lo Studio dei Tumori​​​‌ Dino Amadori - IRST​ SRL (IRST), Italy
    • Centre​‌ Hospitalier Universitaire de Liege​​ (CHUL), Belgium
    • Turkiye Cumhuriyeti​​​‌ Saglik Bakanligi (MOH), Türkiye​
    • Universidad de Vigo (UVIGO),​‌ Spain
    • Arteevo Technologies LTD​​ (ARTEEVO), Israel
  • Inria contact:​​​‌
    Jan Ramon
  • Coordinator:
  • Summary:​

    In recent years, Federated​‌ Learning (FL) has emerged​​ as a revolutionary privacy-enhancing​​​‌ technology and, consequently, has​ quickly expanded to other​‌ applications.

    However, further research​​ has cast a shadow​​​‌ of doubt on the​ strength of privacy protection​‌ provided by FL. Potential​​ vulnerabilities and threats pointed​​​‌ out by researchers included​ a curious aggregator threat;​‌ susceptibility to man-in-the-middle and​​ insider attacks that disrupt​​​‌ the convergence of global​ and local models or​‌ cause convergence to fake​​ minima; and, most importantly,​​​‌ inference attacks that aim​ to re-identify data subjects​‌ from FL’s AI model​​ parameter updates.

    The goal​​​‌ of TRUMPET is to​ research and develop novel​‌ privacy enhancement methods for​​ Federated Learning, and to​​​‌ deliver a highly scalable​ Federated AI service platform​‌ for researchers, that will​​ enable AI-powered studies of​​​‌ siloed, multi-site, cross-domain, cross​ border European datasets with​‌ privacy guarantees that exceed​​ the requirements of GDPR.​​​‌ The generic TRUMPET platform​ will be piloted, demonstrated​‌ and validated in the​​ specific use case of​​​‌ European cancer hospitals, allowing​ researchers and policymakers to​‌ extract AI-driven insights from​​ previously inaccessible cross-border, cross-organization​​ cancer data, while ensuring​​​‌ the patients’ privacy. The‌ strong privacy protection accorded‌​‌ by the platform will​​ be verified through the​​​‌ engagement of external experts‌ for independent privacy leakage‌​‌ and re-identification testing.

    A​​ secondary goal is to​​​‌ research, develop and promote‌ with EU data protection‌​‌ authorities a novel metric​​ and tool for the​​​‌ certification of GDPR compliance‌ of FL implementations.

    The‌​‌ consortium is composed of​​ 9 interdisciplinary partners: 3​​​‌ Research Organizations, 1 University,‌ 3 SMEs and 2‌​‌ Clinical partners with extensive​​ experience and expertise to​​​‌ guarantee the correct performance‌ of the activities and‌​‌ the achievement of the​​ results.

FLUTE

Participant: Jan​​​‌ Ramon [contact person].‌

  • Title:
    Federate Learning and‌​‌ mUlti-party computation Techniques for​​ prostatE cancer
  • Duration:
    From​​​‌ May 1, 2023 to‌ April 30, 2026
  • Partners:‌​‌
    • INRIA, France
    • Quibim Sociedad​​ Limitada (QUIBIM), Spain
    • TIMELEX,​​​‌ Belgium
    • Technovative Solutions LTD,‌ United Kingdom
    • HL7 Europe‌​‌ Foundation, Belgium
    • Fundacion Centro​​ Tecnoloxico de Telecomunicacions de​​​‌ Galicia (GRADIANT), Spain
    • Siemens‌ SRL, Romania
    • Universitat Politecnica‌​‌ de Catalunya (UPC), Spain​​
    • Istituto Romagnolo per lo​​​‌ Studio dei Tumori Dino‌ Amadori - IRST SRL‌​‌ (IRST), Italy
    • Centre Hospitalier​​ Universitaire de Liege (CHUL),​​​‌ Belgium
    • Fundacio Hospital Universitari‌ Vall d'Hebron - Institut‌​‌ de Recerca (VHIR), Spain​​
    • Arteevo Technologies LTD (ARTEEVO),​​​‌ Israel
  • Inria contact:
    Jan‌ Ramon
  • Coordinator:
  • Summary:

    The‌​‌ FLUTE project will advance​​ and scale up data-driven​​​‌ healthcare by developing novel‌ methods for privacy-preserving cross-border‌​‌ utilization of data hubs.​​ Advanced research will be​​​‌ performed to push the‌ performance envelope of secure‌​‌ multi-party computation in Federated​​ Learning, including the associated​​​‌ AI models and secure‌ execution environments. The technical‌​‌ innovations will be integrated​​ in a privacy-enforcing platform​​​‌ that will provide innovators‌ with a provenly secure‌​‌ environment for federated healthcare​​ AI solution development, testing​​​‌ and deployment, including the‌ integration of real world‌​‌ health data from the​​ data hubs and the​​​‌ generation and utilization of‌ synthetic data. To maximize‌​‌ the impact, adoption and​​ replicability of the results,​​​‌ the project will contribute‌ to the global HL7‌​‌ FHIR standard development, and​​ create novel guidelines for​​​‌ GDPR-compliant cross-border Federated Learning‌ in healthcare.

    To demonstrate‌​‌ the practical use and​​ impact of the results,​​​‌ the project will integrate‌ the FLUTE platform with‌​‌ health data hubs located​​ in three different countries,​​​‌ use their data to‌ develop a novel federated‌​‌ AI toolset for diagnosis​​ of clinically significant prostate​​​‌ cancer and perform a‌ multi-national clinical validation of‌​‌ its efficacy, which will​​ help to improve predictions​​​‌ of aggressive prostate cancer‌ while avoiding unnecessary biopsies,‌​‌ thus improving the welfare​​ of patients and significantly​​​‌ reducing the associated costs.‌

    Team. The 11-strong consortium‌​‌ will include three clinical​​ / data partners from​​​‌ three different countries, three‌ technology SMEs, three technology‌​‌ research partners, a legal/ethics​​ partner and a standards​​​‌ organization.

    Collaboration. In accordance‌ with the priorities set‌​‌ by the European Commission,​​ the project will target​​​‌ collaboration, cross-fertilization and synergies‌ with related national and‌​‌ international European projects.

10.3​​ National initiatives

10.3.1 ANR​​​‌ PMR (2020-2025)

Participants: Jan‌ Ramon [contact person],‌​‌ Marc Tommasi.

Given​​​‌ the growing awareness of​ privacy risks of data​‌ processing, there is an​​ increasing interest in privacy-preserving​​​‌ learning. However, shortcomings in​ the state of the​‌ art limit the applicability​​ of the privacy-preserving learning​​​‌ paradigm. First, most approaches​ assume too optimistically a​‌ honest-but-curious setting. Second, most​​ approaches consider one learning​​​‌ task in isolation, not​ accounting for the context​‌ where querying is a​​ recurring activity. We will​​​‌ investigate new algorithms and​ models that address these​‌ shortcomings. Among others, (i)​​ our algorithms will combine​​​‌ privacy-preserving properties of differential​ privacy with security offered​‌ by cryptography and (ii)​​ based on models of​​​‌ information flows in integrated​ data handling processes, we​‌ will build more refined​​ models analyzing the implications​​​‌ of repeated querying. We​ will demonstrate the utility​‌ of our new theory​​ and algorithms by proposing​​​‌ strategies to realistically apply​ them in significant real-world​‌ problems illustrated through use​​ cases in the medical​​​‌ domain

10.3.2 FedMalin. INRIA​ Defi (2021-2026)

Participants: Jan​‌ Ramon, Marc Tommasi​​ [contact person], Michaël​​​‌ Perrot, Batiste Le​ Bars, Edwige Cyffers​‌, Brahim Erraji,​​ Luis Lugo, Paul​​​‌ Andrey.

In many​ use-cases of Machine Learning​‌ (ML), data is naturally​​ decentralized: medical data is​​​‌ collected and stored by​ different hospitals, crowdsensed data​‌ is generated by personal​​ devices, etc. Federated Learning​​​‌ (FL) has recently emerged​ as a novel paradigm​‌ where a set of​​ entities with local datasets​​​‌ collaboratively train ML models​ while keeping their data​‌ decentralized.

FedMalin is a​​ research project that spans​​​‌ 10 Inria research teams​ and aims to push​‌ FL research and concrete​​ use-cases through a multidisciplinary​​​‌ consortium involving expertise in​ ML, distributed systems, privacy​‌ and security, networks, and​​ medicine. We propose to​​​‌ address a number of​ challenges that arise when​‌ FL is deployed over​​ the Internet, including privacy​​​‌ and fairness, energy consumption,​ personalization, and location/time dependencies.​‌

FedMalin will also contribute​​ to the development of​​​‌ open-source tools for FL​ experimentation and real-world deployments,​‌ and use them for​​ concrete applications in medicine​​​‌ and crowdsensing.

10.3.3 COMANCHE:​ Computational Models of Lexical​‌ Meaning and Change. INRIA​​ Action Exploratoire (2022-2026)

Participants:​​​‌ Pascal Denis [contact person]​, Mikaela Keller,​‌ Bastien Liétard, Nassim​​ Boudjenah.

Comanche proposes​​​‌ to transfer and adapt​ recent Natural Language representation​‌ learning algorithms from deep​​ learning to model the​​​‌ evolution of the meaning​ of words, and to​‌ confront these computational models​​ to theories on language​​​‌ acquisition and the diachrony​ of languages. At the​‌ crossroads between machine learning,​​ psycholinguistics and historical linguistics,​​​‌ this project will make​ it possible to validate​‌ or revise some of​​ these theories, but also​​​‌ to bring out computational​ models that are more​‌ sober in terms of​​ data and computations because​​​‌ they exploit new inductive​ biases inspired by these​‌ disciplines.

In collaboration with​​ UMR SCALAB (CNRS, Université​​​‌ de Lille), l’Unité de​ Recherche STIH (Sorbonne Université),​‌ et l’UMR ATILF (CNRS,​​ Université de Lorraine).

10.3.4​​​‌ IPoP, Projet interdisciplinaire sur​ la protection des données​‌ personnelles, PEPR Cybersécurité (2022-2028).​​

Participants: Jan Ramon,​​ Marc Tommasi [contact person]​​​‌, Michaël Perrot,‌ Raouf Kerkouche, Batiste‌​‌ Le Bars, Paul​​ Andrey, Jean Dufraiche​​​‌, Shreya Venugopal.‌

Digital technologies provide services‌​‌ which can greatly increase​​ quality of life (e.g.​​​‌ connected e-health devices, location‌ based services, or personal‌​‌ assistants). However, these services​​ can also raise major​​​‌ privacy risks, as they‌ involve personal data, or‌​‌ even sensitive data. Indeed,​​ this notion of personal​​​‌ data is the cornerstone‌ of French and European‌​‌ regulations, since processing such​​ data triggers a series​​​‌ of obligations that the‌ data controller must abide‌​‌ by. This raises many​​ multidisciplinary issues, as the​​​‌ challenges are not only‌ technological, but also societal,‌​‌ judiciary, economic, political and​​ ethical.

The objectives of​​​‌ this project are thus‌ to study the threats‌​‌ on privacy that have​​ been introduced by these​​​‌ new services, and to‌ conceive theoretical and technical‌​‌ privacy-preserving solutions that are​​ compatible with French and​​​‌ European regulations, that preserve‌ the quality of experience‌​‌ of the users. These​​ solutions will be deployed​​​‌ and assessed, both on‌ the technological and legal‌​‌ sides, and on their​​ societal acceptability. In order​​​‌ to achieve these objectives,‌ we adopt an interdisciplinary‌​‌ approach, bringing together many​​ diverse fields: computer science,​​​‌ technology, engineering, social sciences,‌ economy and law.

The‌​‌ project's scientific program focuses​​ on new forms of​​​‌ personal information collection, on‌ Artificial Intelligence (AI) and‌​‌ its governance, data anonymization​​ techniques, personal data management​​​‌ and distributed calculation protocol‌ privacy preserving infrastructures, differential‌​‌ privacy, personal data legal​​ protection and compliance, and​​​‌ all the associated societal‌ and ethical considerations. This‌​‌ unifying interdisciplinary research program​​ brings together internationally recognized​​​‌ research teams (from universities,‌ engineering schools and institutions)‌​‌ working on privacy, and​​ the French Data Protection​​​‌ Authority (CNIL).

This holistic‌ vision of the issues‌​‌ linked to personal data​​ protection will on the​​​‌ one hand let us‌ propose solutions to the‌​‌ scientific and technological challenges​​ and on the other​​​‌ help us confront these‌ solutions in many different‌​‌ ways, in the context​​ of interdisciplinary collaborations, thus​​​‌ leading to recommendations and‌ proposals in the field‌​‌ of regulations or legal​​ frameworks. This comprehensive consideration​​​‌ of all the issues‌ aims at encouraging the‌​‌ adoption and acceptability of​​ the solutions proposed by​​​‌ all stakeholders, legislators, data‌ controllers, data processors, solution‌​‌ designers, developers all the​​ way to end-users.

10.3.5​​​‌ SSF-ML-DH, PEPR Santé Numérique‌ (2022-2028).

Participants: Marc Tommasi‌​‌ [contact person], Batiste​​ Le Bars.

The​​​‌ healthcare sector (public and‌ private) generates an unparalleled‌​‌ amount of data from​​ sources as diverse as​​​‌ electronic medical records, advanced‌ imaging techniques, high-throughput sequencing,‌​‌ wearable devices, and public​​ health data. Leveraging these​​​‌ massive datasets through sophisticated‌ machine learning algorithms has‌​‌ the potential to transform​​ medical practice by enabling​​​‌ the development of more‌ effective and personalized treatments,‌​‌ interventions, and public policies,​​ ultimately improving healthcare delivery​​​‌ and population well-being. However,‌ the highly sensitive nature‌​‌ of health data, cybersecurity​​ risks, biases in the​​​‌ data, and the lack‌ of robustness in machine‌​‌ learning algorithms are key​​​‌ obstacles currently preventing the​ full realization of recent​‌ advancements in artificial intelligence.​​

To overcome these challenges,​​​‌ it is essential to​ address ethical, legal, security,​‌ and robustness issues. This​​ project aims to develop​​​‌ new machine learning algorithms​ that account for the​‌ multi-scale and heterogeneous characteristics​​ of health data while​​​‌ ensuring privacy, robustness against​ adversarial attacks and changes​‌ in data and model​​ dynamics, and fairness for​​​‌ underrepresented populations. By addressing​ these obstacles, we hope​‌ to unlock the barriers​​ that hinder the deployment​​​‌ of innovative solutions in​ digital health.

Specifically, the​‌ project will focus on​​ the following challenges: (i)​​​‌ privacy-preserving learning through differential​ privacy techniques and homomorphic​‌ encryption; (ii) federated learning​​ by balancing accuracy and​​​‌ privacy; (iii) robustness against​ adversarial attacks and changes​‌ in data and model​​ dynamics; (iv) automated “forgetting”​​​‌ mechanisms to implement the​ right to be forgotten.​‌

The project, part of​​ Project of PEPR Digital​​​‌ Health (2023-2027), brings together​ a unique consortium of​‌ experts in machine learning,​​ cybersecurity, statistics, and medical​​​‌ applications. Moreover, it is​ strategically positioned between two​‌ national programs (PEPRs: Cybersecurity​​ and Digital Health), providing​​​‌ a unique opportunity to​ disseminate knowledge and best​‌ practices.

10.3.6 CAPS'UL (2023-2028)​​

Participant: Marc Tommasi [contact​​​‌ person], Paul Andrey​, Simon Decomble.​‌

The project is built​​ around 3 axes.

  1. Promote​​​‌ a common digital health​ culture among all current​‌ and future healthcare professionals:​​ cybersecurity issues, legal and​​​‌ ethical regulation of healthcare​ data, communication and digital​‌ health tools, telehealth framework.​​
  2. Design a high-performance tool​​​‌ for practical situations, enabling​ concrete and effective collaboration​‌ between the various training,​​ socio-economic and medico-social players​​​‌ in the implementation of​ training courses. This shared​‌ resource center will provide​​ a credible immersive environment​​​‌ (real software and simulated​ healthcare data) and teaching​‌ scenarios for the entire​​ teaching community. Co-constructed with​​​‌ industry software publishers, it​ will be accessible from​‌ simulation centers and remotely,​​ to meet the different​​​‌ needs of the region.​
  3. Train professionals in the​‌ new digital health support​​ professions, by emphasizing the​​​‌ delivery of “health and​ specific digital issues” courses​‌ that are shared between​​ the various existing courses.​​​‌ These innovative, coherent schemes​ will serve as demonstrators​‌ of excellence on a​​ regional scale.

Magnet will​​​‌ provide tools for synthetic​ data generation with privacy​‌ guarantees dedicated to the​​ immersive environment.

10.3.7 ANR-JCJC​​​‌ FaCTor: Fairness Constraints and​ Guarantees for Trustworthy Machine​‌ Learning (2023-2027)

Participants: Michaël​​ Perrot [contact person],​​​‌ Marc Tommasi, Shreya​ Venugopal.

The goal​‌ of the FaCTor project​​ is to provide ML​​​‌ practitioners with theoretically well​ founded means to develop​‌ algorithms that come with​​ fairness guarantees. It points​​​‌ toward the development of​ trustworthy and socially acceptable​‌ ML solutions. The end​​ goal is to make​​​‌ the models more accountable​ and in line with​‌ the requirements of the​​ law, ensuring that the​​​‌ benefits of ML are​ not limited to a​‌ subset of the population.​​

10.3.8 REDEEM: Resilient, Decentralized​​​‌ and Privacy-Preserving Machine Learning,​ PEPR IA (2022-2028).

Participants:​‌ Jan Ramon [contact person]​​, Marc Tommasi,​​ Michaël Perrot, Arnaud​​​‌ Descours, Batiste Le‌ Bars, Raouf Kerkouche‌​‌.

The vision of​​ distributed AI is attractive​​​‌ because it contributes to‌ user empowerment by limiting‌​‌ the dissemination of personal​​ and confidential information to​​​‌ a single node in‌ the network and it‌​‌ makes systems independent of​​ a superior force that​​​‌ would decide what is‌ good for everyone. But‌​‌ on the other hand​​ it opens up major​​​‌ issues of security and‌ robustness: how can we‌​‌ guarantee the compliance of​​ a model learned in​​​‌ another context? How can‌ we protect our AI‌​‌ network from the introduction​​ of biased knowledge, malicious​​​‌ or not, or even‌ “backdoor” functions? If the‌​‌ pooling consists of a​​ simultaneous optimisation, how can​​​‌ we ensure the validity‌ of contributions that are‌​‌ not always explicable?

The​​ action led on the​​​‌ theme of distributed AI‌ is therefore at the‌​‌ confluence of the topics​​ Embedded and Frugality (distributed​​​‌ systems are frequently low-resource‌ embedded systems such as‌​‌ telephones, vehicles or autonomous​​ robots) and Trust, as​​​‌ the issues of security,‌ reliability and robustness are‌​‌ shed in a new​​ light in collaborative AI.​​​‌

The REDEEM project brings‌ together a consortium of‌​‌ complementary teams and researchers,​​ with primary expertise in​​​‌ machine learning, distributed optimization,‌ consensus algorithms and game‌​‌ theory. It also associates​​ a unique spectrum of​​​‌ research orientation, from highly‌ theoretical work on convergence‌​‌ of distributed learning algorithms​​ to extensive experiences towards​​​‌ practical and efficient implementations‌ as well as innovative‌​‌ dissemination activities.

10.3.9 ANR-JCJC​​ Adada: Adada: Adaptive Datasets​​​‌ for Enhancing Reasoning in‌ Large Language Models (2024-2028)‌​‌

Participants: Damien Sileo [contact​​ person], Pascal Denis​​​‌.

Large Language Models‌ (LLMs) are neural networks‌​‌ designed for text completion,​​ playing a pivotal role​​​‌ in various Natural Language‌ Processing (NLP) applications such‌​‌ as conversational assistance and​​ document analysis. Given the​​​‌ widening scope of LLM‌ applications, generating truly useful‌​‌ completions goes beyond mere​​ linguistic fluency. It requires​​​‌ logical precision, multi-step reasoning‌ abilities, and adherence to‌​‌ userdefined constraints. These capabilities​​ are essential for tackling​​​‌ the implicit reasoning tasks‌ woven into everyday scenarios,‌​‌ from interpreting texts with​​ embedded rules to evaluating​​​‌ products against technical specifications‌ or identifying inconsistencies. At‌​‌ their core, such tasks​​ involve complex logical problems​​​‌ intertwined with the nuances‌ of natural language and‌​‌ with background knowledge. Logical​​ reasoning remains challenging for​​​‌ current LLMs. As a‌ countermeasure, we can train‌​‌ neural models to mimic​​ the output of symbolic​​​‌ reasoning systems (e.g., logic‌ theorem provers, or other‌​‌ algorithms) on procedurally generated​​ problems, like Q which​​​‌ actually comes from the‌ RuleTaker dataset, to sharpen‌​‌ their reasoning capabilities. This​​ training improves accuracy on​​​‌ human-authored problems. However synthetic‌ problem datasets are currently‌​‌ generated once, sometimes not​​ reproducibly. They can quickly​​​‌ become too easy for‌ ongoing models after being‌​‌ included in the training​​ data or due to​​​‌ model scaling. Adada proposes‌ a novel framework to‌​‌ distill modern symbolic reasoning​​ into language models through​​​‌ evolutive synthetic datasets. By‌ explicitly steering problem generation‌​‌ to improve a specific​​​‌ model on a targeted​ downstream task, Adada seeks​‌ to continuously enhance language​​ models for reasoning-intensive applications​​​‌ such as technical documentation​ understanding, commonsense reasoning, and​‌ legal analysis.

10.3.10 ANR​​ Melissa: MEthodological contributions in​​​‌ statistical Learning InSpired by​ SurfAce engineering (2025-2029)

Participants:​‌ Marc Tommasi [contact person]​​, Batiste Le Bars​​​‌, Rémi Gilleron,​ Jan Ramon.

The​‌ underlying dynamics of many​​ physical problems are governed​​​‌ by parameterized partial differential​ equations (PDEs). Despite important​‌ scientific advances in numerical​​ simulation, solving efficiently PDEs​​​‌ remains complex and often​ prohibitively expensive. Physics-informed Machine​‌ Learning (PiML) has recently​​ emerged as a promising​​​‌ way to learn efficient​ surrogate solvers, and augment​‌ the physical laws by​​ leveraging knowledge extracted from​​​‌ data. From a machine​ learning perspective, ignoring the​‌ fundamental principles of the​​ underlying physics may lead​​​‌ to ill-posed problems and​ thus to implausible solutions​‌ yielding poor generalization.

Numerous​​ algorithmic contributions in deep-learning​​​‌ have recently exploited domain​ knowledge for (i) designing​‌ suitable physics-regularized loss functions,​​ (ii) initializing neural networks​​​‌ with meaningful parameters, (iii)​ guiding the design of​‌ consistent architectures, or (iv)​​ building hybrid models.

Despite​​​‌ indisputable advances, PiML remains​ an emerging topic with​‌ several open problems that​​ remain to be addressed:​​​‌ (i) Deriving generalization/approximation guarantees;​ (ii) Learning with a​‌ limited amount of data;​​ (iii) Augmenting partially known​​​‌ physical laws; (v) Modeling​ uncertainty; (vi) Building foundation​‌ models for physics.

Developing​​ suited solutions that tackle​​​‌ these interrelated challenges is​ crucial for the usability​‌ of PiML in realistic​​ scenarios. This is the​​​‌ goal of MELISSA which​ gathers 3 teams (Inria​‌ MALICE, Inria MAGNET and​​ MLIA) with a strong​​​‌ expertise at the interface​ of machine learning, optimization​‌ and physics. By conducting​​ this project from both​​​‌ theoretical and algorithmic perspectives,​ the objective is to​‌ design the next generation​​ of provably accurate PIML​​​‌ algorithms in the challenging​ context of laser-matter interaction​‌ where data is scarce​​ and the available physical​​​‌ laws only partially explain​ the observed dynamics.

10.4​‌ Regional initiatives

10.4.1 Cross​​ Disciplinary Project (CDP) Prime​​​‌ Next Gen: NEXT-GENeration PRecIsion​ medicine in Inflammatory and​‌ MEtabolic diseases (2026-2030)

Participants:​​ Marc Tommasi [contact person]​​​‌, Jan Ramon,​ Michaël Perrot.

This​‌ project has been accepted​​ in 2025 and will​​​‌ start in 2026.

The​ overall objective of PRIME​‌ NEXT-GEN is to define,​​ characterize and validate precise​​​‌ endotypes of patients with​ metabolic diseases and IMIDs,​‌ towards precision medicine. The​​ scientific hypotheses are that​​​‌ 1. the level of​ meta-inflammation will be key​‌ in defining these endotypes​​ 2. studying metabolic diseases​​​‌ and IMIDs together could​ provide a new clinically​‌ relevant pathophysiological nosography of​​ these associated diseases. The​​​‌ further innovative approach developed​ in this project is​‌ to validate these endotypes​​ at all pathophysiological, temporal​​​‌ and acceptability levels.

10.4.2​ Cross Disciplinary Project (CDP)​‌ LOOP: closed Loop neurOtechnologies:​​ from sensOrs to aPplications​​​‌ (2026-2030)

Participants: Marc Tommasi​ [contact person], Jan​‌ Ramon.

This project​​ has been accepted in​​​‌ 2025 and will start​ in 2026.

the project​‌ will implement an interdisciplinary​​ strategy applied to the​​ development of a mechanism-based​​​‌ therapy for drug-resistant hallucinations‌ in patients with schizophrenia.‌​‌ The project will be​​ structured around three main​​​‌ research pillars: 1) the‌ algorithmic level of closed-loop‌​‌ systems (online signal processing,​​ machine learning); 2) the​​​‌ design of a complete‌ and non-invasive solution to‌​‌ address the question of​​ refractory hallucinations, and study​​​‌ how closed-loop interaction schemes‌ can be beneficial for‌​‌ highly impaired psychiatric patients;​​ 3) the design of​​​‌ the sensors themselves and‌ translational research with animal‌​‌ models, in order to​​ enhance the quality of​​​‌ the recordings and pave‌ the way for future‌​‌ invasive miniaturized solutions. Finally,​​ in addition to these​​​‌ three vertical research axes,‌ this CDP also seeks‌​‌ to offer a unique​​ opportunity to examine the​​​‌ development of neurotechnologies in‌ the accelerating context of‌​‌ regulations at the European​​ level, with law and​​​‌ ethics forming the first‌ cross-cutting axis.

11 Dissemination‌​‌

11.1 Promoting scientific activities​​

11.1.1 Scientific events: selection​​​‌

  • Marc Tommasi served as‌ Area Chair for UAI‌​‌ and ICLR, PC member​​ of ECML, APVP and​​​‌ CAp.
  • Jan Ramon served‌ as reviewer for AAAI,‌​‌ AISTATS, ECML-PKDD, ICLR, ICML,​​ IJCAI, NeurIPS, SDM, UAI​​​‌ and Icbinb@ICLR.
  • Damien Sileo‌ served as a reviewer‌​‌ for ACL.
  • Michaël Perrot​​ served as reviewer for​​​‌ ICML, NeurIPS, AISTATS, and‌ CAp.
  • Batiste Le Bars‌​‌ served as a reviewer​​ for NeurIPS, AISTATS and​​​‌ ICLR
  • Pascal Denis was‌ Action Editor for ACL‌​‌ Rolling Review, and served​​ as PC member for​​​‌ COLING 2025
  • Mikaela Keller‌ served as a reviewer‌​‌ for ACL, AISTATS and​​ CAp

11.1.2 Journal

Member​​​‌ of editorial boards
  • Jan‌ Ramon is member of‌​‌ the editorial boards of​​ Machine Learning Journal (MLJ),​​​‌ Data Mining and Knowledge‌ Discovery (DMKD).
  • Pascal Denis‌​‌ standing reviewer for Transactions​​ of the Association for​​​‌ Computational Linguistics (TACL).
Reviewer‌ - reviewing activities
  • Batiste‌​‌ Le Bars served as​​ reviewer for the Electronic​​​‌ Journal of Statistics (EJS)‌ and the SIAM Journal‌​‌ on Optimization (SIOPT)

11.1.3​​ Invited talks

  • Jan Ramon​​​‌ gave presentations at IPEAC‌ Secured/FLUTE workshop (Barcelona, 1/25);‌​‌ HS Booster final event​​ (Brussels, 5/3/25); Data health​​​‌ summit (Brussels, 20/3/25); IHI-HaDEA‌ meeting on synthetic data‌​‌ (online) : Synthetic data​​ in the FLUTE project​​​‌ (03/25); pepr-ipop workshop on‌ legal aspects of AI‌​‌ (Paris, 20/3/25); Privacy symposium​​ (Venice, 20/3/25); FHIN-FLUTE meeting​​​‌ (online): introduction to FLUTE‌ for the FHIN (www.fhin.be)‌​‌ consortium (05/25); EBDVF (Copenhagen);​​ OncoLille (Lille, 24/11/25); CRIStAL​​​‌ axe IA (Lille, 19/12/25).‌
  • Marc Tommasi gave presentations‌​‌ at FedMalin (Rennes, 02/25);​​ the prospective INRIA seminar​​​‌ (Rungis, 03/25); RedChainLab workshop‌ (Lyon, 06/25).
  • Michaël Perrot‌​‌ gave a presentation in​​ the SIGMA Team (Lille,​​​‌ 05/25).
  • Damien Sileo gave‌ a presentation at the‌​‌ LLM4Roq lecture group (Paris,​​ 17/10/25)
  • Batiste Le Bars​​​‌ gave a presentation in‌ the PreMeDICaL Team (Montpellier,‌​‌ 04/25); at the ANR​​ Melissa Kick-off meeting (Saint-Etienne,​​​‌ 05/25); and during the‌ Journées de Statistique de‌​‌ la SFdS (Marseille, 06/25)​​
  • Raouf Kerkouche gave a​​​‌ presentation at PrivateAIM on‌ Differentially private federated learning‌​‌ for localized control of​​ infectious disease dynamics (Germany,​​​‌ 10/25)

11.1.4 Leadership within‌ the scientific community

  • Jan‌​‌ Ramon was member of​​​‌ the bureau of the​ Societé Savante Francophone d'Apprentissage​‌ Machine (SSFAM)
  • Pascal Denis​​ is co-head of the​​​‌ CNRS GDR "Langues et​ langage à la croisée​‌ des disciplines" (LLcD) and​​ co-animator of the Working​​​‌ Group NLP & Cognition​ of the CNRS GDR-TAL.​‌

11.1.5 Scientific expertise

  • Marc​​ Tommasi was a member​​​‌ (scientific expert) of the​ recruitment committee of full​‌ professors at Saint-Etienne and​​ Lille.
  • Marc Tommasi was​​​‌ a member (scientific expert)​ of the recruitment committee​‌ of associate professors at​​ Lille.
  • Marc Tommasi was​​​‌ expert for CY Generation​
  • Marc Tommasi was member​‌ of the ANR CE​​ 39, Sécurité Globale.
  • Jan​​​‌ Ramon was reviewer for​ COST project proposals.
  • Pascal​‌ Denis was a member​​ (scientific expert) of the​​​‌ CRCN/ISFP recruitment committee at​ INRIA Center of Bordeaux​‌ University, as well as​​ acting parity and equal​​​‌ opportunities co-officer on that​ committee.
  • Mikaela Keller was​‌ a member (scientific expert)​​ of recruitment committees of​​​‌ assistant professors in Besançon​ and Bordeaux

11.1.6 Research​‌ administration

  • Mikaela Keller was​​ a vice-president of CER​​​‌ (Commission Emploi Recherche) in​ the INRIA Center of​‌ Lille University and a​​ facilitator of the CRIStAL-wide​​​‌ AI Axis that promotes​ discussion among the CRIStAL​‌ teams working on AI.​​
  • Marc Tommasi is co-head​​​‌ of the DatInG group​ (3 teams, about 80​‌ persons), member of the​​ Conseil Scientifique du laboratoire​​​‌ CRIStAL and member of​ the Commission mixte CRIStAL/Faculty​‌ of Science, Lille University.​​ He is member of​​​‌ the BCEP (bureau du​ comité des équipes projet).​‌
  • Pascal Denis is also​​ a member of the​​​‌ network "référents données" at​ Inria and Université de​‌ Lille (Lille Open Research​​ Data). He is also​​​‌ administrator of Inria membership​ to Linguistic Data Consortium​‌ (LDC). He is also​​ member of the local​​​‌ sustainable development committee (CLDD)​ at INRIA Center of​‌ Lille University.
  • Michaël Perrot​​ is a substitue member​​​‌ of the Comité de​ Centre (named, representing the​‌ administration). Michaël Perrot is​​ the local correspondent for​​​‌ Activity Reports in the​ Inria center at the​‌ university of Lille. Michaël​​ Perrot is volunteer in​​​‌ the local AGOS team.​
  • Jan Ramon is a​‌ member of the Comité​​ de Centre and member​​​‌ of the bureau of​ the SSFAM (Société Savante​‌ Francophone d'Apprentissage Machine)
  • Damien​​ Sileo is a member​​​‌ of the Comité de​ l'Evaluation de l'IA.
  • Batiste​‌ Le Bars is in​​ charge of the Magnet​​​‌ Seminar organization

11.2 Teaching​ - Supervision - Juries​‌ - Educational and pedagogical​​ outreach

11.2.1 Teaching

  • Licence​​​‌ MIASHS: Mikaela Keller ,​ Data Science, 24h, L2,​‌ Université de Lille.
  • Licence​​ Informatique: Marc Tommasi ,​​​‌ Introduction to AI, 24h,​ L2, Université de Lille.​‌
  • Master Computer Science: Mikaela​​ Keller , Apprentissage profond,​​​‌ 24h, M1, Université de​ Lille.
  • Master Computer Science:​‌ Mikaela Keller , Machine​​ learning pour le traitement​​​‌ automatique du language naturel,​ 24h, M2, Université de​‌ Lille.
  • Master Computer Science:​​ Marc Tommasi , Data​​​‌ Science, 48h, M1, Université​ de Lille.
  • Master Computer​‌ Science: Batiste Le Bars​​ , Data Science, 36h,​​​‌ M1, Université de Lille.​
  • Master Data Science: Marc​‌ Tommasi Seminars 24h.
  • Master​​ Data Science: Damien Sileo​​ , Natural Language Processing,​​​‌ 24h, M2, Université de‌ Lille et Ecole Centrale‌​‌ de Lille.
  • Master Data​​ Science: Damien Sileo ,​​​‌ Natural Language Processing, 4.5h,‌ M2, Institut Mines-Télécom.
  • Master‌​‌ Data Science: Michaël Perrot​​ , Fairness in Trustworthy​​​‌ Machine Learning, 24h, M2,‌ Université de Lille et‌​‌ Ecole Centrale de Lille.​​
  • Master Cognitive Sciences: Michaël​​​‌ Perrot , Fairness in‌ Machine Leanring, 6h, M2,‌​‌ Université de Lille.

11.2.2​​ Supervision

  • Postdoc: Arnaud Descours​​​‌ . On federated optimization‌ with lower communication cost.‌​‌ Since Nov. 2023. Jan​​ Ramon
  • Postdoc: Luis Lugo​​​‌ . On federated learning‌ with energy budgets, since‌​‌ Jun. 24. Marc Tommasi​​ and Romain Rouvoy .​​​‌
  • Postdoc: Jean-Baptiste Fermanian .‌ On Personalized Federated Learning,‌​‌ since Oct. 25. Aurélien​​ Bellet and Batiste Le​​​‌ Bars .
  • Phd defended:‌ Marc Damie . Secure‌​‌ protocols for verifiable decentralized​​ machine learning. Jan Ramon​​​‌ with Andreas Peter (U.‌ Twente, NL & U.‌​‌ Oldenburg, DE) Florian Hahn​​ (University of Twente, NL).​​​‌40
  • Phd defended: Dinh‌ Viet-Toan Le . Natural‌​‌ Language Processing approaches in​​ the musical domain :​​​‌ suitability, performance and limits.‌ Mikaela Keller and Louis‌​‌ Bigo41
  • PhD defended:​​ Antoine Barczewski . Transparent​​​‌ privacy-preserving machine learning. Jan‌ Ramon .39
  • PhD‌​‌ in progress: Bastien Liétard​​ . Computational Models of​​​‌ Lexical Semantic Change, since‌ Nov. 2022. Anne Carlier‌​‌ (Université Paris Sorbonne), Pascal​​ Denis and Mikaela Keller​​​‌
  • PhD in progress: Aleksei‌ Korneev . Trustworthy multi-site‌​‌ privacy-enhancingtechnologies, since Dec. 2022.​​ Jan Ramon
  • PhD in​​​‌ progress: Clément Pierquin .‌ Synthetic data generation with‌​‌ privacy constraints, since Sept.​​ 2023, Aurélien Bellet and​​​‌ Marc Tommasi
  • PhD in‌ progress: Gabriel Loiseau Transfert‌​‌ and multitask learning approaches​​ for text anonymizaton, since​​​‌ Sept. 2023, Damien Sileo‌ and Marc Tommasi
  • PhD‌​‌ in progress: Brahim Erraji​​ . Fairness in Federated​​​‌ Learning, since Sept. 2023,‌ Aurélien Bellet , Catuscia‌​‌ Palamidessi and Michaël Perrot​​ .
  • PhD in progress:​​​‌ Shreya Venugopal . Guaranteed‌ Fairness in Machine Learning,‌​‌ since Oct. 24, Michaël​​ Perrot .
  • PhD in​​​‌ progress: Jean Dufraiche .‌ Fairness and Privacy in‌​‌ Machine Learning, since Oct.​​ 24, Michaël Perrot ,​​​‌ Marc Tommasi .
  • PhD‌ in progress: Thomas Boudou‌​‌ . Private and Byzantine-Robust​​ Federated Learning, since Oct.​​​‌ 24, Aurélien Bellet and‌ Batiste Le Bars
  • PhD‌​‌ in progress: Paul Andrey​​ , Synthetic data and​​​‌ privacy, since Nov 24.‌ Marc Tommasi and Batiste‌​‌ Le Bars
  • PhD in​​ progress: Quentin Sinh .​​​‌ Towards a generic, decentralized,‌ secure and privacy preserving‌​‌ automated learning, Jul 24,​​ Jan Ramon
  • PhD in​​​‌ progress: Dimitri Kachler .‌ Methodologies for the Generation,‌​‌ Analysis and Selection of​​ Procedurally Generated Data for​​​‌ the Training and Evaluation‌ of Large Language Models,‌​‌ since Nov. 25, Damien​​ Sileo and Pascal Denis​​​‌
  • PhD in progress: Nassim‌ Boudjenah . Computational Models‌​‌ of Semantic Memory, since​​ Dec. 25, Pascal Denis​​​‌ and Rémi Gilleron
  • Engineer:‌ Jules Boulet , FLUTE‌​‌ and TRUMPET projects, until​​ May 25, Jan Ramon​​​‌
  • Engineer: Elina Thibeau Sutre‌ , FLUTE and TRUMPET‌​‌ projects until June 25,​​ Jan Ramon
  • Engineer: Jules​​​‌ Yvon , FLUTE and‌ TRUMPET projects, since Sep.‌​‌ 24, Jan Ramon
  • Engineer:​​​‌ Younes Ikli , FLUTE​ and TRUMPET projects, since​‌ Sep. 24, Jan Ramon​​
  • Engineer: Léonard Deroose .​​​‌ Development of TRUMPET privacy-preserving​ platform components, Since Sep.​‌ 2023. Jan Ramon .​​
  • Engineer: Baptiste Cottier .​​​‌ Zero-knowledge verification of computations,​ since Jan. 2025. Jan​‌ Ramon .
  • Engineer: Valentin​​ Lacombe . Enhancing the​​​‌ core reasoning capabilities of​ LLMs with RLVR (Reinforcement​‌ Learning with Verifiable Reward),​​ leveraging a modular problem​​​‌ generation library, since Feb.​ 2025. Damien Sileo .​‌
  • Engineer: Zakaria El Bouchouari​​ , Privacy preserving decentralized​​​‌ learning prototypes for FLUTE​ and REDEEM, since Feb.​‌ 25, Jan Ramon
  • Engineer:​​ Alexandre Louvet , differential​​​‌ privacy and the security​ of multi-party computation in​‌ the FLUTE and TRUMPET​​ projects, since Sept. 25,​​​‌ Jan Ramon
  • Engineer: Simon​ Decomble . Development of​‌ synthetic data generators for​​ the CAPS'UL project, since​​​‌ Apr. 2025. Jan Ramon​ .

11.2.3 Juries

  • Marc​‌ Tommasi was member of​​ the following PhD juries:​​​‌ Volodimir Mitarchuk (Rapporteur), Marianne​ Abi Kanaan (Rapporteur), Yacine​‌ Belal (Examinateur), Louis Roussel​​ (President), Viet-Toan Le (Directeur),​​​‌ Jade Garcia-Bourrée (Rapporteur), Pierre​ Jobic (Rapporteur).
  • Marc Tommasi​‌ was president of the​​ HDR of Charlotte Laclau.​​​‌

11.2.4 Educational and pedagogical​ outreach

  • Marc Tommasi is​‌ directeur des études for​​ the Machine Learning master​​​‌ of Computer Science.

11.3​ Popularization

11.3.1 Participation in​‌ Live events

  • Michaël Perrot​​ gave an introductory presentation​​​‌ to machine learning during​ a Journée du Numérique​‌ on “Intelligence artificielle :​​ enjeux éducatifs et pistes​​​‌ pédagogiques” organized by INSPÉ​ Lille HdF (Arras, 03/25).​‌

11.3.2 Others science outreach​​ relevant activities

  • Marc Tommasi​​​‌ gave a presentation on​ responsible AI at the​‌ "séminaire national des secrétaires​​ généraux d'académies et de​​​‌ région académiques" (Roubaix, 6/25).​
  • Michaël Perrot lead the​‌ scientific team advising the​​ participants of the Serious​​​‌ Game Jam on Artificial​ Intelligence organized by INSPÉ​‌ Lille HdF and the​​ Université de Lille (Lille,​​​‌ 01/25).
  • Michaël Perrot gave​ an introductory presentation to​‌ fairness in machine learning​​ to SNT and NSI​​​‌ teachers for the Académie​ de Lille (Lille, 06/25).​‌
  • Damien Sileo was interviewed​​ by Episloon
  • Michaël Perrot​​​‌ participated to a round​ table on artificial intelligence​‌ during the Semaine NSI​​ (Lille, 12/25).
  • Batiste Le​​​‌ Bars gave a poster​ presentation at the AI​‌ Action Summit Conference (École​​ Polytechnique, 02/25)
  • Mikaela Keller​​​‌ co-organized with Mathieu Giraud​ from Algomus, CRIStAL, a​‌ one day seminar journée​​ de recherche en musiques​​​‌ to encourage collaborations between​ several disciplines working with​‌ Music in Université de​​ Lille (Lille, 03/25)
  • Mikaela​​​‌ Keller participated to a​ round table with jurists​‌ on AI for research​​ in law during Séminaire​​​‌ doctoral du CRDP (Lille,​ 07/25)

12 Scientific production​‌

12.1 Major publications

  • 1​​ inproceedingsA.Aurélien Bellet​​​‌, R.Rachid Guerraoui​ and H.Hadrien Hendrikx​‌. Who started this​​ rumor? Quantifying the natural​​​‌ differential privacy guarantees of​ gossip protocols.DISC​‌ 2020 - 34th International​​ Symposium on Distributed Computing​​​‌Freiburg / Virtual, Germany​InriaOctober 2020HAL​‌
  • 2 inproceedingsA.Aurélien​​ Bellet, R.Rachid​​​‌ Guerraoui, M.Mahsa​ Taziki and M.Marc​‌ Tommasi. Personalized and​​ Private Peer-to-Peer Machine Learning​​.AISTATS 2018 -​​​‌ 21st International Conference on‌ Artificial Intelligence and Statistics‌​‌Lanzarote, SpainApril 2018​​, 1-20HAL
  • 3​​​‌ inproceedingsM.Mathieu Dehouck‌ and P.Pascal Denis‌​‌. Delexicalized Word Embeddings​​ for Cross-lingual Dependency Parsing​​​‌.EACL1EACL‌ 2017Valencia, SpainApril‌​‌ 2017, 241 -​​ 250HALDOI
  • 4​​​‌ inproceedingsM.Mathieu Dehouck‌ and P.Pascal Denis‌​‌. Phylogenetic Multi-Lingual Dependency​​ Parsing.NAACL 2019​​​‌ - Annual Conference of‌ the North American Chapter‌​‌ of the Association for​​ Computational LinguisticsProceedings of​​​‌ the 2019 Conference of‌ the North American Chapter‌​‌ of the Association for​​ Computational Linguistics: Human Language​​​‌ TechnologiesMinneapolis, United States‌June 2019HAL
  • 5‌​‌ articleP.Peter Kairouz​​, B. H.Brendan​​​‌ H. McMahan, B.‌Brendan Avent, A.‌​‌Aurélien Bellet, M.​​Mehdi Bennis, A.​​​‌ N.Arjun Nitin Bhagoji‌, K.Kallista Bonawitz‌​‌, Z.Zachary Charles​​, G.Graham Cormode​​​‌, R.Rachel Cummings‌, R. G.Rafael‌​‌ Gregorio Lucas D'Oliveira,​​ S. E.Salim El​​​‌ Rouayheb, D.David‌ Evans, J.Josh‌​‌ Gardner, Z.Zachary​​ Garrett, A.Adrià​​​‌ Gascón, B.Badih‌ Ghazi, P. .‌​‌Phillip B. Gibbons,​​ M.Marco Gruteser,​​​‌ Z.Zaid Harchaoui,‌ C.Chaoyang He,‌​‌ L.Lie He,​​ Z.Zhouyuan Huo,​​​‌ B.Ben Hutchinson,‌ J.Justin Hsu,‌​‌ M.Martin Jaggi,​​ T.Tara Javidi,​​​‌ G.Gauri Joshi,‌ M.Mikhail Khodak,‌​‌ J.Jakub Konečný,​​ A.Aleksandra Korolova,​​​‌ F.Farinaz Koushanfar,‌ S.Sanmi Koyejo,‌​‌ T.Tancrède Lepoint,​​ Y.Yang Liu,​​​‌ P.Prateek Mittal,‌ M.Mehryar Mohri,‌​‌ R.Richard Nock,​​ A.Ayfer Ozgür,​​​‌ R.Rasmus Pagh,‌ M.Mariana Raykova,‌​‌ H.Hang Qi,​​ D.Daniel Ramage,​​​‌ R.Ramesh Raskar,‌ D.Dawn Song,‌​‌ W.Weikang Song,​​ S. .Sebastian Urban​​​‌ Stich, Z.Ziteng‌ Sun, A. T.‌​‌Ananda Theertha Suresh,​​ F.Florian Tramèr,​​​‌ P.Praneeth Vepakomma,‌ J.Jianyu Wang,‌​‌ L.Li Xiong,​​ Z.Zheng Xu,​​​‌ Q.Qiang Yang,‌ F. X.Felix X.‌​‌ Yu, H.Han​​ Yu and S.Sen​​​‌ Zhao. Advances and‌ Open Problems in Federated‌​‌ Learning.Foundations and​​ Trends in Machine Learning​​​‌141-22021,‌ 1-210HAL
  • 6 inproceedings‌​‌O.Ondřej Kużelka,​​ Y.Yuyi Wang and​​​‌ J.Jan Ramon.‌ Bounds for Learning from‌​‌ Evolutionary-Related Data in the​​ Realizable Case.International​​​‌ Joint Conference on Artificial‌ Intelligence (IJCAI)Proceedings of‌​‌ the International Joint Conference​​ on Artificial Intelligence (IJCAI)​​​‌ 2016New York, United‌ StatesJuly 2016HAL‌​‌
  • 7 inproceedingsE.Emmanuel​​ Lassalle and P.Pascal​​​‌ Denis. Joint Anaphoricity‌ Detection and Coreference Resolution‌​‌ with Constrained Latent Structures​​.AAAI Conference on​​​‌ Artificial Intelligence (AAAI 2015)‌Proceedings of the Twenty-Ninth‌​‌ AAAI Conference on Artificial​​ Intelligence (AAAI 2015)Austin,​​​‌ Texas, United StatesJanuary‌ 2015HAL
  • 8 inproceedings‌​‌B.Batiste Le Bars​​​‌, A.Aurélien Bellet​, M.Marc Tommasi​‌, K.Kevin Scaman​​ and G.Giovanni Neglia​​​‌. Improved Stability and​ Generalization Guarantees of the​‌ Decentralized SGD Algorithm.​​ICML 2024 - The​​​‌ Forty-first International Conference on​ Machine LearningVienne, Austria​‌July 2024HAL
  • 9​​ articleS.Shayne Longpre​​​‌, R.Robert Mahari​, A.Anthony Chen​‌, N.Naana Obeng-Marnu​​, D.Damien Sileo​​​‌, W.William Brannon​, N.Niklas Muennighoff​‌, N.Nathan Khazam​​, J.Jad Kabbara​​​‌, K.Kartik Perisetla​, X.Xinyi Wu​‌, E.Enrico Shippole​​, K.Kurt Bollacker​​​‌, T.Tongshuang Wu​, L.Luis Villa​‌, S.Sandy Pentland​​ and S.Sara Hooker​​​‌. A large-scale audit​ of dataset licensing and​‌ attribution in AI.​​Nature Machine Intelligence6​​​‌8August 2024,​ 975-987HALDOI
  • 10​‌ inproceedingsG.Gaurav Maheshwari​​, P.Pascal Denis​​​‌, M.Mikaela Keller​ and A.Aurélien Bellet​‌. Fair NLP Models​​ with Differentially Private Text​​​‌ Encoders.Findings of​ the Association for Computational​‌ Linguistics: EMNLP 2022Abu​​ Dhabi, United Arab Emirates​​​‌2022HAL
  • 11 inproceedings​P.Paul Mangold,​‌ M.Michaël Perrot,​​ A.Aurélien Bellet and​​​‌ M.Marc Tommasi.​ Differential Privacy has Bounded​‌ Impact on Fairness in​​ Classification.Proceedings of​​​‌ the 40th International Conference​ on Machine LearningInternational​‌ Conference on Machine Learning​​202Honolulu, United States​​​‌July 2023HAL
  • 12​ articleC.Christos Pelekis​‌, J.Jan Ramon​​ and Y.Yuyi Wang​​​‌. Holder-type inequalities​ and their applications to​‌ concentration and correlation bounds​​.Indagationes Mathematicae28​​​‌12017, 170--182​HALDOI
  • 13 article​‌C.César Sabater,​​ A.Aurélien Bellet and​​​‌ J.Jan Ramon.​ An Accurate, Scalable and​‌ Verifiable Protocol for Federated​​ Differentially Private Averaging.​​​‌Machine LearningOctober 2022​HALDOI
  • 14 article​‌A. S.Ali Shahin​​ Shamsabadi, B. M.​​​‌Brij Mohan Lal Srivastava​, A.Aurélien Bellet​‌, N.Nathalie Vauquier​​, E.Emmanuel Vincent​​​‌, M.Mohamed Maouche​, M.Marc Tommasi​‌ and N.Nicolas Papernot​​. Differentially private speaker​​​‌ anonymization.Proceedings on​ Privacy Enhancing Technologies2023​‌1January 2023HAL​​
  • 15 inproceedingsB. M.​​​‌Brij Mohan Lal Srivastava​, N.Nathalie Vauquier​‌, M.Md Sahidullah​​, A.Aurélien Bellet​​​‌, M.Marc Tommasi​ and E.Emmanuel Vincent​‌. Evaluating Voice Conversion-based​​ Privacy Protection against Informed​​​‌ Attackers.ICASSP 2020​ - 45th International Conference​‌ on Acoustics, Speech, and​​ Signal ProcessingIEEE Signal​​​‌ Processing SocietyBarcelona, Spain​May 2020, 2802-2806​‌HAL
  • 16 inproceedingsP.​​Paul Vanhaesebrouck, A.​​​‌Aurélien Bellet and M.​Marc Tommasi. Decentralized​‌ Collaborative Learning of Personalized​​ Models over Networks.​​​‌International Conference on Artificial​ Intelligence and Statistics (AISTATS)​‌Fort Lauderdale, Florida., United​​ StatesApril 2017HAL​​​‌

12.2 Publications of the​ year

International journals

  • 17​‌ articleM.Mélodie Bellegarda​​, G.Gary Boddaert​​​‌, S.Sophie Dufour​, D.Dominique Knutsen​‌ and A.Angele Brunelliere​​. Neural evidence for​​ perceiving a vowel merger​​​‌ after a social interaction‌ within a native language‌​‌.Brain and Language​​261February 2025,​​​‌ 105529HALDOIback‌ to text
  • 18 article‌​‌C.Cécile Bossaert,​​ S.Sébastien Volant,​​​‌ P.Paul Andrey,‌ V.Valery Hedouin and‌​‌ E.Eric Wiel.​​ Medicolegal obstacles in pre-hospital​​​‌ care: An overview of‌ practice in the North‌​‌ of France in 2023​​.Archives of Legal​​​‌ Medicine164December‌ 2025, 200623HAL‌​‌DOI
  • 19 articleA.​​Alicia Fasquel, W.​​​‌Wassila El Mardi,‌ I.Isabelle Bonnotte,‌​‌ D.Dominique Knutsen and​​ A.Angèle Brunellière.​​​‌ How does the creation‌ of new semantic relationships‌​‌ during dialogue impact long-term​​ semantic representations after dialogue?​​​‌Acta Psychologica260September‌ 2025, 105513HAL‌​‌DOIback to text​​
  • 20 articleA.Aleksei​​​‌ Korneev and J.Jan‌ Ramon. A Survey‌​‌ on Verifiable Cross-Silo Federated​​ Learning.Transactions on​​​‌ Machine Learning Research Journal‌June 2025HALback‌​‌ to text
  • 21 article​​D.-V.Dinh-Viet-Toan Le,​​​‌ L.Louis Bigo,‌ D.Dorien Herremans and‌​‌ M.Mikaela Keller.​​ Natural Language Processing Methods​​​‌ for Symbolic Music Generation‌ and Information Retrieval: a‌​‌ Survey.ACM Computing​​ Surveys577January​​​‌ 2025, 1-40HAL‌DOIback to text‌​‌
  • 22 articleT.Thibaud​​ Leteno, M.Michael​​​‌ Perrot, C.Charlotte‌ Laclau, A.Antoine‌​‌ Gourru and C.Christophe​​ Gravier. Fair Text​​​‌ Classification via Transferable Representations‌.Journal of Machine‌​‌ Learning Research26239​​December 2025, 1--47​​​‌HALback to text‌
  • 23 articleL.Long‌​‌ Phan, A.Alice​​ Gatti, Z.Ziwen​​​‌ Han, N.Nathaniel‌ Li, J.Josephina‌​‌ Hu, H.Hugh​​ Zhang, S.Sean​​​‌ Shi, M.Michael‌ Choi, A.Anish‌​‌ Agrawal, A.Arnav​​ Chopra, A.Adam​​​‌ Khoja, R.Ryan‌ Kim, J.Jason‌​‌ Hausenloy, O.Oliver​​ Zhang, M.Mantas​​​‌ Mazeika, D.Daron‌ Anderson, T.Tung‌​‌ Nguyen, M.Mobeen​​ Mahmood, F.Fiona​​​‌ Feng, S. Y.‌Steven Y. Feng,‌​‌ H.Haoran Zhao,​​ M.Michael Yu,​​​‌ V.Varun Gangal,‌ C.Chelsea Zou,‌​‌ Z.Zihan Wang,​​ J. P.Jessica P.​​​‌ Wang, P.Pawan‌ Kumar, O.Oleksandr‌​‌ Pokutnyi, R.Robert​​ Gerbicz, S.Serguei​​​‌ Popov, J.-C.John-Clark‌ Levin, M.Mstyslav‌​‌ Kazakov, J.Johannes​​ Schmitt, G.Geoff​​​‌ Galgon, A.Alvaro‌ Sanchez, Y.Yongki‌​‌ Lee, W.Will​​ Yeadon, S.Scott​​​‌ Sauers, M.Marc‌ Roth, C.Chidozie‌​‌ Agu, S.Søren​​ Riis, F.Fabian​​​‌ Giska, S.Saiteja‌ Utpala, Z.Zachary‌​‌ Giboney, G. M.​​Gashaw M. Goshu,​​​‌ J. o.Joan of‌ Arc Xavier, S.-J.‌​‌Sarah-Jane Crowson, M.​​ M.Mohinder Maheshbhai Naiya​​​‌, N.Noah Burns‌, L.Lennart Finke‌​‌, Z.Zerui Cheng​​, H.Hyunwoo Park​​​‌, F.Francesco Fournier-Facio‌, J.John Wydallis‌​‌, M.Mark Nandor​​​‌, A.Ankit Singh​, T.Tim Gehrunger​‌, J.Jiaqi Cai​​, B.Ben Mccarty​​​‌, D.Darling Duclosel​, J.Jungbae Nam​‌, J.Jennifer Zampese​​, R. G.Ryan​​​‌ G. Hoerr, A.​Aras Bacho, G.​‌ A.Gautier Abou Loume​​, A.Abdallah Galal​​​‌, H.Hangrui Cao​, A. C.Alexis​‌ C Garretson, D.​​Damien Sileo, Q.​​​‌Qiuyu Ren, D.​Doru Cojoc, P.​‌Pavel Arkhipov, U.​​Usman Qazi, L.​​​‌Lianghui Li, S.​Sumeet Motwani, C.​‌ S.Christian Schroeder de​​ Witt, E.Edwin​​​‌ Taylor, J.Johannes​ Veith, E.Eric​‌ Singer, T. D.​​Taylor D. Hartman,​​​‌ P.Paolo Rissone,​ J.Jaehyeok Jin,​‌ J. W.Jack Wei​​ Lun Shi, C.​​​‌ G.Chris G. Willcocks​, J.Joshua Robinson​‌, A.Aleksandar Mikov​​, A.Ameya Prabhu​​​‌, L.Longke Tang​, X.Xavier Alapont​‌, J. L.Justine​​ Leon Uro, K.​​​‌Kevin Zhou, E.​ d.Emily de Oliveira​‌ Santos, A. P.​​Andrey Pupasov Maksimov,​​​‌ E.Edward Vendrow,​ K.Kengo Zenitani,​‌ J.Julien Guillod,​​ Y.Yuqi Li,​​​‌ J.Joshua Vendrow,​ V.Vladyslav Kuchkin,​‌ N.Ng Ze-An,​​ P.Pierre Marion,​​​‌ D.Denis Efremov,​ J.Jayson Lynch,​‌ K.Kaiqu Liang,​​ A.Andrew Gritsevskiy,​​​‌ D.Dakotah Martinez,​ B.Ben Pageler,​‌ N.Nick Crispino,​​ D.Dimitri Zvonkine,​​​‌ N. W.Natanael Wildner​ Fraga, S.Saeed​‌ Soori, O.Ori​​ Press, H.Henry​​​‌ Tang, J.Julian​ Salazar, S. R.​‌Sean R. Green,​​ L.Lina Brüssel,​​​‌ M.Moon Twayana,​ A.Aymeric Dieuleveut,​‌ T. R.T. Ryan​​ Rogers, W.Wenjin​​​‌ Zhang, B.Bikun​ Li, J.Jinzhou​‌ Yang, A.Arun​​ Rao, G.Gabriel​​​‌ Loiseau, M.Mikhail​ Kalinin, M.Marco​‌ Lukas, C.Ciprian​​ Manolescu, S.Subrata​​​‌ Mishra, A. G.​Ariel Ghislain Kemogne Kamdoum​‌, T.Tobias Kreiman​​, T.Tad Hogg​​​‌, A.Alvin Jin​, C.Carlo Bosio​‌, G.Gongbo Sun​​, B. P.Brian​​​‌ P Coppola, T.​Tim Tarver, H.​‌Haline Heidinger, R.​​Rafael Sayous, S.​​​‌Stefan Ivanov, J.​ M.Joseph M Cavanagh​‌, J.Jiawei Shen​​, J. M.Joseph​​​‌ Marvin Imperial, P.​Philippe Schwaller, S.​‌Shaipranesh Senthilkuma, A.​​ M.Andres M Bran​​​‌, A.Ali Dehghan​, A.Andres Algaba​‌, B.Brecht Verbeken​​, D.David Noever​​​‌, R.Ragavendran P​ V, L.Lisa​‌ Schut, I.Ilia​​ Sucholutsky, E.Evgenii​​​‌ Zheltonozhskii, D.Derek​ Lim, R.Richard​‌ Stanley, S.Shankar​​ Sivarajan, T.Tong​​​‌ Yang, J.John​ Maar, J.Julian​‌ Wykowski, M.Martí​​ Oller, J.Jennifer​​​‌ Sandlin, A.Anmol​ Sahu, Y.Yuzheng​‌ Hu, S.Sara​​ Fish, N.Nasser​​ Heydari, A.Archimedes​​​‌ Apronti, K.Kaivalya‌ Rawal, T. G.‌​‌Tobias Garcia Vilchis,​​ Y.Yuexuan Zu,​​​‌ M.Martin Lackner,‌ J.James Koppel,‌​‌ J.Jeremy Nguyen,​​ D. S.Daniil S.​​​‌ Antonenko, S.Steffi‌ Chern, B.Bingchen‌​‌ Zhao, P.Pierrot​​ Arsene, A.Alan​​​‌ Goldfarb, S.Sergey‌ Ivanov, R.Rafał‌​‌ Poświata, C.Chenguang​​ Wang, D.Daofeng​​​‌ Li, D.Donato‌ Crisostomi, A.Andrea‌​‌ Achilleos, B.Benjamin​​ Myklebust, A.Archan​​​‌ Sen, D.David‌ Perrella, N.Nurdin‌​‌ Kaparov, M. H.​​Mark H Inlow,​​​‌ A.Allen Zang,‌ E.Elliott Thornley,‌​‌ D.Daniil Orel,​​ V.Vladislav Poritski,​​​‌ S.Shalev Ben-David,‌ Z.Zachary Berger,‌​‌ P.Parker Whitfill,​​ M.Michael Foster,​​​‌ D.Daniel Munro,‌ L.Linh Ho,‌​‌ D. B.Dan Bar​​ Hava, A.Aleksey​​​‌ Kuchkin, R.Robert‌ Lauff, D.David‌​‌ Holmes, F.Frank​​ Sommerhage, K.Keith​​​‌ Schneider, Z.Zakayo‌ Kazibwe, N.Nate‌​‌ Stambaugh, M.Mukhwinder​​ Singh, I.Ilias​​​‌ Magoulas, D.Don‌ Clarke, D. H.‌​‌Dae Hyun Kim,​​ F. M.Felipe Meneguitti​​​‌ Dias, V.Veit‌ Elser, K. P.‌​‌Kanu Priya Agarwal,​​ V. E.Victor Efren​​​‌ Guadarrama Vilchis, I.‌Immo Klose, C.‌​‌Christoph Demian, U.​​Ujjwala Anantheswaran, A.​​​‌Adam Zweiger, G.‌Guglielmo Albani, J.‌​‌Jeffery Li, N.​​Nicolas Daans, M.​​​‌Maksim Radionov, V.‌Václav Rozhoň, Z.‌​‌Ziqiao Ma, C.​​Christian Stump, M.​​​‌Mohammed Berkani, J.‌Jacob Platnick, V.‌​‌Volodymyr Nevirkovets, L.​​Luke Basler, M.​​​‌Marco Piccardo, F.‌Ferenc Jeanplong, N.‌​‌Niv Cohen, J.​​Josef Tkadlec, P.​​​‌Paul Rosu, P.‌Piotr Padlewski, S.‌​‌Stanislaw Barzowski, K.​​Kyle Montgomery, A.​​​‌Aline Menezes, A.‌Arkil Patel, Z.‌​‌Zixuan Wang, J.​​Jamie Tucker-Foltz, J.​​​‌Jack Stade, T.‌Tom Goertzen, F.‌​‌Fereshteh Kazemi, J.​​Jeremiah Milbauer, J.​​​‌ A.John Arnold Ambay‌, A.Abhishek Shukla‌​‌, Y. C.Yan​​ Carlos Leyva Labrador,​​​‌ A.Alan Givré,‌ H.Hew Wolff,‌​‌ V.Vivien Rossbach,​​ M. F.Muhammad Fayez​​​‌ Aziz, Y.Younesse‌ Kaddar, Y.Yanxu‌​‌ Chen, R.Robin​​ Zhang, J.Jiayi​​​‌ Pan, A.Antonio‌ Terpin, N.Niklas‌​‌ Muennighoff, H.Hailey​​ Schoelkopf, E.Eric​​​‌ Zheng, A.Avishy‌ Carmi, A.Adam‌​‌ Jones, J.Jainam​​ Shah, E. D.​​​‌Ethan D. L. Brown‌, K.Kelin Zhu‌​‌, M.Max Bartolo​​, R.Richard Wheeler​​​‌, A.Andrew Ho‌, S.Shaul Barkan‌​‌, J.Jiaqi Wang​​, M.Martin Stehberger​​​‌, E.Egor Kretov‌, K.Kaustubh Sridhar‌​‌, Z.Zienab El-Wasif​​, A.Anji Zhang​​​‌, D.Daniel Pyda‌, J.Joanna Tam‌​‌, D. M.David​​​‌ M. Cunningham, V.​Vladimir Goryachev, D.​‌Demosthenes Patramanis, M.​​Michael Krause, A.​​​‌Andrew Redenti, D.​Daniel Bugas, D.​‌David Aldous, J.​​Jesyin Lai, S.​​​‌Shannon Coleman, M.​Mohsen Bahaloo, J.​‌Jiangnan Xu, S.​​Sangwon Lee, S.​​​‌Sandy Zhao, N.​Ning Tang, M.​‌ K.Michael K. Cohen​​, M.Micah Carroll​​​‌, O.Orr Paradise​, J. H.Jan​‌ Hendrik Kirchner, S.​​Stefan Steinerberger, M.​​​‌Maksym Ovchynnikov, J.​ O.Jason O. Matos​‌, A.Adithya Shenoy​​, B. A.Benedito​​​‌ Alves de Oliveira Junior​, M.Michael Wang​‌, Y.Yuzhou Nie​​, P.Paolo Giordano​​​‌, P.Philipp Petersen​, A.Anna Sztyber-Betley​‌, P.Priti Shukla​​, J.Jonathan Crozier​​​‌, A.Antonella Pinto​, S.Shreyas Verma​‌, P.Prashant Joshi​​, Z.-X.Zheng-Xin Yong​​​‌, A.Allison Tee​, J.Jérémy Andréoletti​‌, O.Orion Weller​​, R.Raghav Singhal​​​‌, G.Gang Zhang​, A.Alexander Ivanov​‌, S.Seri Khoury​​, H.Hamid Mostaghimi​​​‌, K.Kunvar Thaman​, Q.Qijia Chen​‌, T. Q.Tran​​ Quoc Khánh, J.​​​‌Jacob Loader, S.​Stefano Cavalleri, H.​‌Hannah Szlyk, Z.​​Zachary Brown, J.​​​‌Jonathan Roberts, W.​William Alley, K.​‌Kunyang Sun, R.​​Ryan Stendall, M.​​​‌Max Lamparth, A.​Anka Reuel, T.​‌Ting Wang, H.​​Hanmeng Xu, S.​​​‌ G.Sreenivas Goud Raparthi​, P.Pablo Hernández-Cámara​‌, F.Freddie Martin​​, D.Dmitry Malishev​​​‌, T.Thomas Preu​, T.Tomek Korbak​‌, M.Marcus Abramovitch​​, D.Dominic Williamson​​​‌, Z.Ziye Chen​, B.Biró Bálint​‌, M. S.M​​ Saiful Bari, P.​​​‌Peyman Kassani, Z.​Zihao Wang, B.​‌Behzad Ansarinejad, L.​​ P.Laxman Prasad Goswami​​​‌, Y.Yewen Sun​, H.Hossam Elgnainy​‌, D.Daniel Tordera​​, G.George Balabanian​​​‌, E.Earth Anderson​, L.Lynna Kvistad​‌, A. J.Alejandro​​ José Moyano, R.​​​‌Rajat Maheshwari, A.​Ahmad Sakor, M.​‌Murat Eron, I.​​ C.Isaac C. Mcalister​​​‌, J.Javier Gimenez​, I.Innocent Enyekwe​‌, A. F.Andrew​​ Favre D. O.,​​​‌ S.Shailesh Shah,​ X.Xiaoxiang Zhou,​‌ F.Firuz Kamalov,​​ R.Ronald Clark,​​​‌ S.Sherwin Abdoli,​ T.Tim Santens,​‌ K.Khalida Meer,​​ H. K.Harrison K​​​‌ Wang, K.Kalyan​ Ramakrishnan, E.Evan​‌ Chen, A.Alessandro​​ Tomasiello, G. B.​​​‌G. Bruno de Luca​, S.-Z.Shi-Zhuo Looi​‌, V.-K.Vinh-Kha Le​​, N.Noam Kolt​​​‌, N.Niels Mündler​, A.Avi Semler​‌, E.Emma Rodman​​, J.Jacob Drori​​​‌, C. J.Carl​ J Fossum, M.​‌Milind Jagota, R.​​Ronak Pradeep, H.​​​‌Honglu Fan, T.​Tej Shah, J.​‌Jonathan Eicher, M.​​Michael Chen, K.​​Kushal Thaman, W.​​​‌William Merrill, C.‌Carter Harris, J.‌​‌Jason Gross, I.​​Ilya Gusev, A.​​​‌Asankhaya Sharma, S.‌Shashank Agnihotri, P.‌​‌Pavel Zhelnov, S.​​Siranut Usawasutsakorn, M.​​​‌Mohammadreza Mofayezi, S.‌Sergei Bogdanov, A.‌​‌Alexander Piperski, M.​​Marc Carauleanu, D.​​​‌ K.David K. Zhang‌, D.Dylan Ler‌​‌, R.Roman Leventov​​, I.Ignat Soroko​​​‌, T.Thorben Jansen‌, P.Pascal Lauer‌​‌, J.Joshua Duersch​​, V.Vage Taamazyan​​​‌, W.Wiktor Morak‌, W.Wenjie Ma‌​‌, W.William Held​​, T. Đ.Tran​​​‌ Đuc Huy, R.‌Ruicheng Xian, A.‌​‌ R.Armel Randy Zebaze​​, M.Mohanad Mohamed​​​‌, J. N.Julian‌ Noah Leser, M.‌​‌ X.Michelle X Yuan​​, L.Laila Yacar​​​‌, J.Johannes Lengler‌, H.Hossein Shahrtash‌​‌, E.Edson Oliveira​​, J. W.Joseph​​​‌ W. Jackson, D.‌ E.Daniel Espinosa Gonzalez‌​‌, A.Andy Zou​​, M.Muthu Chidambaram​​​‌, T.Timothy Manik‌, H.Hector Haffenden‌​‌, D.Dashiell Stander​​, A.Ali Dasouqi​​​‌, A.Alexander Shen‌, E.Emilien Duc‌​‌, B.Bita Golshani​​, D.David Stap​​​‌, M.Mikalai Uzhou‌, A. B.Alina‌​‌ Borisovna Zhidkovskaya, L.​​Lukas Lewark, M.​​​‌Mátyás Vincze, D.‌Dustin Wehr, C.‌​‌Colin Tang, Z.​​Zaki Hossain, S.​​​‌Shaun Phillips, J.‌Jiang Muzhen, F.‌​‌Fredrik Ekström, A.​​Angela Hammon, O.​​​‌Oam Patel, N.‌Nicolas Remy, F.‌​‌Faraz Farhidi, G.​​George Medley, F.​​​‌Forough Mohammadzadeh, M.‌Madellene Peñaflor, H.‌​‌Haile Kassahun, A.​​Alena Friedrich, C.​​​‌Claire Sparrow, T.‌Taom Sakal, O.‌​‌Omkar Dhamane, A.​​ K.Ali Khajegili Mirabadi​​​‌, E.Eric Hallman‌, M.Mike Battaglia‌​‌, M.Mohammad Maghsoudimehrabani​​, H.Hieu Hoang​​​‌, A.Alon Amit‌, D.Dave Hulbert‌​‌, R.Roberto Pereira​​, S.Simon Weber​​​‌, S.Stephen Mensah‌, N.Nathan Andre‌​‌, A.Anton Peristyy​​, C.Chris Harjadi​​​‌, H.Himanshu Gupta‌, S.Stephen Malina‌​‌, S.Samuel Albanie​​, W.Will Cai​​​‌, M.Mustafa Mehkary‌, F.Frank Reidegeld‌​‌, A.-K.Anna-Katharina Dick​​, C.Cary Friday​​​‌, J.Jasdeep Sidhu‌, W.Wanyoung Kim‌​‌, M.Mariana Costa​​, H.Hubeyb Gurdogan​​​‌, B.Brian Weber‌, H.Harsh Kumar‌​‌, T.Tong Jiang​​, A.Arunim Agarwal​​​‌, C.Chiara Ceconello‌, W. S.Warren‌​‌ S. Vaz, C.​​Chao Zhuang, H.​​​‌Haon Park, A.‌ R.Andrew R. Tawfeek‌​‌, D.Daattavya Aggarwal​​, M.Michael Kirchhof​​​‌, L.Linjie Dai‌, E.Evan Kim‌​‌, J.Johan Ferret​​, Y.Yuzhou Wang​​​‌, M.Minghao Yan‌, K.Krzysztof Burdzy‌​‌, L.Lixin Zhang​​, A.Antonio Franca​​​‌, D. T.Diana‌ T. Pham, K.‌​‌ Y.Kang Yong Loh​​​‌, J.Joshua Robinson​, S.Shreen Gul​‌, G.Gunjan Chhablani​​, Z.Zhehang Du​​​‌, A.Adrian Cosma​, C.Colin White​‌, R.Robin Riblet​​, P.Prajvi Saxena​​​‌, J.Jacob Votava​, V.Vladimir Vinnikov​‌, E.Ethan Delaney​​, S.Shiv Halasyamani​​​‌, S. M.Syed​ M. Shahid, J.-C.​‌Jean-Christophe Mourrat, L.​​Lavr Vetoshkin, R.​​​‌Renas Bacho, V.​Vincent Ginis, A.​‌Aleksandr Maksapetyan, F.​​Florencia de la Rosa​​​‌, X.Xiuyu Li​, G.Guillaume Malod​‌, L.Leon Lang​​, J.Julien Laurendeau​​​‌, F.Fatimah Adesanya​, J.Julien Portier​‌, L.Lawrence Hollom​​, V.Victor Souza​​​‌, Y. A.Yuchen​ Anna Zhou, Y.​‌Yiğit Yalın, G.​​ D.Gbenga Daniel Obikoya​​​‌, L.Luca Arnaboldi​, F.Filippo Bigi​‌, K.Kaniuar Bacho​​, P.Pierre Clavier​​​‌, G.Gabriel Recchia​, M.Mara Popescu​‌, N.Nikita Shulga​​, N. M.Ngefor​​​‌ Mildred Tanwie, T.​ C.Thomas C. H.​‌ Lux, B.Ben​​ Rank, C.Colin​​​‌ Ni, A.Alesia​ Yakimchyk, H.Huanxu​‌ Liu, O.Olle​​ Häggström, E.Emil​​​‌ Verkama, H.Himanshu​ Narayan, H.Hans​‌ Gundlach, L.Leonor​​ Brito-Santana, B.Brian​​​‌ Amaro, V.Vivek​ Vajipey, R.Rynaa​‌ Grover, Y.Yiyang​​ Fan, G. P.​​​‌Gabriel Poesia Reis E​ Silva, L.Linwei​‌ Xin, Y.Yosi​​ Kratish, J.Jakub​​​‌ Łucki, W.-D.Wen-Ding​ Li, J.Justin​‌ Xu, K. J.​​Kevin Joseph Scaria,​​​‌ F.Freddie Vargus,​ F.Farzad Habibi,​‌ E.Emanuele Rodolà,​​ J.Jules Robins,​​​‌ V.Vincent Cheng,​ D.Declan Grabb,​‌ I.Ida Bosio,​​ T.Tony Fruhauff,​​​‌ I.Ido Akov,​ E. J.Eve J.​‌ Y. Lo, H.​​Hao Qi, X.​​​‌Xi Jiang, B.​Ben Segev, J.​‌Jingxuan Fan, S.​​Sarah Martinson, E.​​​‌ Y.Erik Y. Wang​, K.Kaylie Hausknecht​‌, M. P.Michael​​ P. Brenner, M.​​​‌Mao Mao, Y.​Yibo Jiang, X.​‌Xinyu Zhang, D.​​David Avagian, E.​​​‌ J.Eshawn Jessica Scipio​, M. R.Muhammad​‌ Rehan Siddiqi, A.​​Alon Ragoler, J.​​​‌Justin Tan, D.​Deepakkumar Patil, R.​‌Rebeka Plecnik, A.​​Aaron Kirtland, R.​​​‌ G.Roselynn Grace Montecillo​, S.Stephane Durand​‌, O. F.Omer​​ Faruk Bodur, Z.​​​‌Zahra Adoul, M.​Mohamed Zekry, G.​‌Guillaume Douville, A.​​Ali Karakoc, T.​​​‌ C.Tania C. B.​ Santos, S.Samir​‌ Shamseldeen, L.Loukmane​​ Karim, A.Anna​​​‌ Liakhovitskaia, N.Nate​ Resman, N.Nicholas​‌ Farina, J. C.​​Juan Carlos Gonzalez,​​​‌ G.Gabe Maayan,​ S.Sarah Hoback,​‌ R. d.Rodrigo de​​ Oliveira Pena, G.​​​‌Glen Sherman, H.​Hodjat Mariji, R.​‌Rasoul Pouriamanesh, W.​​Wentao Wu, G.​​Gözdenur Demir, S.​​​‌Sandra Mendoza, I.‌Ismail Alarab, J.‌​‌Joshua Cole, D.​​Danyelle Ferreira, B.​​​‌Bryan Johnson, H.‌Hsiaoyun Milliron, M.‌​‌Mohammad Safdari, L.​​Liangti Dai, S.​​​‌Siriphan Arthornthurasuk, A.‌Alexey Pronin, J.‌​‌Jing Fan, A.​​Angel Ramirez-Trinidad, A.​​​‌Ashley Cartwright, D.‌Daphiny Pottmaier, O.‌​‌Omid Taheri, D.​​David Outevsky, S.​​​‌Stanley Stepanic, S.‌Samuel Perry, L.‌​‌Luke Askew, R.​​ A.Raúl Adrián Huerta​​​‌ Rodríguez, A.Abdelkader‌ Dendane, S.Sam‌​‌ Ali, R.Ricardo​​ Lorena, K.Krishnamurthy​​​‌ Iyer, S. M.‌Sk Md Salauddin,‌​‌ M.Murat Islam,​​ J.Juan Gonzalez,​​​‌ J.Josh Ducey,‌ R.Russell Campbell,‌​‌ M.Maja Somrak,​​ V.Vasilios Mavroudis,​​​‌ E.Eric Vergo,‌ J.Juehang Qin,‌​‌ B.Benjámin Borbás,​​ E.Eric Chu,​​​‌ J.Jack Lindsey,‌ A.Anil Radhakrishnan,‌​‌ A.Antoine Jallon,​​ I. M.I. M.​​​‌ J. Mcinnis, A.‌Alex Hoover, S.‌​‌Sören Möller, S.​​Song Bian, J.​​​‌John Lai, T.‌Tejal Patwardhan, S.‌​‌Summer Yue, A.​​Alexandr Wang and D.​​​‌Dan Hendrycks. A‌ benchmark of expert-level academic‌​‌ questions to assess AI​​ capabilities.Nature649​​​‌8099January 2026,‌ 1139-1146HALDOIback‌​‌ to text
  • 24 article​​W.Wei Sun,​​​‌ M.Mingxiao Li,‌ D.Damien Sileo,‌​‌ J.Jesse Davis and​​ M.-F.Marie-Francine Moens.​​​‌ Generating Explanations in Medical‌ Question-Answering by Expectation Maximization‌​‌ Inference over Evidence.​​ACM Transactions on Computing​​​‌ for Healthcare62‌2025, 23HAL‌​‌DOIback to text​​

International peer-reviewed conferences

  • 25​​​‌ inproceedingsP.Paul Andrey‌, B.Batiste Le‌​‌ Bars and M.Marc​​ Tommasi. TAMIS: Tailored​​​‌ Membership Inference Attacks on‌ Synthetic Data.Machine‌​‌ Learning and Knowledge Discovery​​ in Databases. Research Track:​​​‌ European Conference, ECML PKDD‌ 2025, Porto, Portugal, September‌​‌ 15–19, 2025, Proceedings, Part​​ VECML PKDD 2025​​​‌ - European Conference on‌ Machine Learning and Principles‌​‌ and Practice of Knowledge​​ Discovery in Databases16017​​​‌Lecture Notes in Computer‌ SciencePorto (Portugal), Portugal‌​‌Springer Nature SwitzerlandSeptember​​ 2025, 203-220HAL​​​‌DOIback to text‌
  • 26 inproceedingsM.Marc‌​‌ Damie and E.Edwige​​ Cyffers. Fedivertex: a​​​‌ Graph Dataset based on‌ Decentralized Social Media.‌​‌WWW '26 - ACM​​ Web Conference 2026Dubai,​​​‌ United Arab Emirates2026‌HALDOIback to‌​‌ text
  • 27 inproceedingsM.​​Marc Damie, J.-B.​​​‌Jean-Benoist Leger, F.‌Florian Hahn and A.‌​‌Andreas Peter. Revisiting​​ the Attacker’s Knowledge in​​​‌ Inference Attacks Against Searchable‌ Symmetric Encryption.23rd‌​‌ International Conference on Applied​​ Cryptography and Network Security​​​‌ (ACNS 2025)15826Munich‌ (Allemagne), GermanySpringer Nature‌​‌ SwitzerlandJune 2025,​​ 370-399HALDOIback​​​‌ to text
  • 28 inproceedings‌B.Bram van Dartel‌​‌, M.Marc Damie​​ and F.Florian Hahn​​​‌. Evaluating Membership Inference‌ Attacks in Heterogeneous-Data Setups‌​‌.Applied Cryptography and​​​‌ Network Security Workshops15655​Lecture Notes in Computer​‌ ScienceMunich, GermanySpringer​​ Nature SwitzerlandOctober 2026​​​‌, 109-117HALDOI​back to text
  • 29​‌ inproceedingsB.Batiste Le​​ Bars and P.Pierre​​​‌ Humbert. On Volume​ Minimization in Conformal Regression​‌.Proceedings of Machine​​ Learning Research.International Conference​​​‌ on Machine Learning (ICML)​Volume 267: International Conference​‌ on Machine Learning, 13-19​​ July 2025, Vancouver Convention​​​‌ Center, Vancouver, CanadaProceedings​ of Machine Learning Research.​‌Vancouver, CanadaJuly 2025​​HALback to text​​​‌
  • 30 inproceedingsD.-V.Dinh-Viet-Toan​ Le and Y.-H.Yi-Hsuan​‌ Yang. METEOR: Melody-aware​​ Texture-controllable Symbolic Orchestral Music​​​‌ Generation via Transformer VAE​.International Joint Conference​‌ on Artificial Intelligence AI,​​ Arts & Creativity (IJCAI​​​‌ 2025)Montreal, CanadaInternational​ Joint Conferences on Artificial​‌ Intelligence Organization2025,​​ 10126-10134HALDOIback​​​‌ to text
  • 31 inproceedings​G.Gabriel Loiseau,​‌ D.Damien Sileo,​​ D.Damien Riquet,​​​‌ M.Maxime Meyer and​ M.Marc Tommasi.​‌ TAROT: Task-Oriented Authorship Obfuscation​​ Using Policy Optimization Methods​​​‌.Proceedings of the​ Sixth Workshop on Privacy​‌ in Natural Language Processing​​Albuquerque, United StatesAssociation​​​‌ for Computational LinguisticsApril​ 2025, 14-31HAL​‌DOIback to text​​
  • 32 inproceedingsG.Gabriel​​​‌ Loiseau, D.Damien​ Sileo, D.Damien​‌ Riquet, M.Maxime​​ Meyer and M.Marc​​​‌ Tommasi. Tau-Eval: A​ Unified Evaluation Framework for​‌ Useful and Private Text​​ Anonymization.Proceedings of​​​‌ the 2025 Conference on​ Empirical Methods in Natural​‌ Language Processing: System Demonstrations​​Suzhou, FranceAssociation for​​​‌ Computational Linguistics2025,​ 216-227HALDOIback​‌ to text
  • 33 inproceedings​​S.Shayne Longpre,​​​‌ N.Nikhil Singh,​ M.Manuel Cherep,​‌ K.Kushagra Tiwary,​​ J.Joanna Materzynska,​​​‌ W.William Brannon,​ R.Robert Mahari,​‌ N.Naana Obeng-Marnu,​​ M.Manan Dey,​​​‌ M.Mohammed Hamdy,​ N.Nayan Saxena,​‌ A.Ahmad Mustafa Anis​​, E.Emad A.​​​‌ Alghamdi, V.Vu​ Minh Chien, D.​‌Da Yin, K.​​Kun Qian, Y.​​​‌Yizhi Li, M.​Minnie Liang, A.​‌An Dinh, S.​​Shrestha Mohanty, D.​​​‌Deividas Mataciunas, T.​Tobin South, J.​‌Jianguo Zhang, A.​​Ariel N. Lee,​​​‌ C.Campbell S. Lund​, C.Christopher Klamm​‌, D.Damien Sileo​​, D.Diganta Misra​​​‌, E.Enrico Shippole​, K.Kevin Klyman​‌, L.Lester James​​ Validad Miranda, N.​​​‌Niklas Muennighoff, S.​Seonghyeon Ye, S.​‌Seungone Kim, V.​​Vipul Gupta, V.​​​‌Vivek Sharma, X.​Xuhui Zhou, C.​‌Caiming Xiong, L.​​Luis Villa, S.​​​‌Stella Biderman, A.​Alex Pentland, S.​‌Sara Hooker and J.​​Jad Kabbara. BRIDGING​​​‌ THE DATA PROVENANCE GAP​ ACROSS TEXT, SPEECH, AND​‌ VIDEO.ICLR 2025​​Singapore (SG), SingaporeApril​​​‌ 2025HALback to​ text
  • 34 inproceedingsC.​‌Clément Pierquin, A.​​Aurélien Bellet, M.​​​‌Marc Tommasi and M.​Matthieu Boussard. Privacy​‌ Amplification Through Synthetic Data:​​ Insights from Linear Regression​​.ICML 2025 -​​​‌ 42nd International Conference on‌ Machine LearningVancouver, Canada‌​‌2025HALDOIback​​ to text
  • 35 inproceedings​​​‌N.Natalia Tomashenko,‌ E.Emmanuel Vincent and‌​‌ M.Marc Tommasi.​​ Analysis of Speech Temporal​​​‌ Dynamics in the Context‌ of Speaker Verification and‌​‌ Voice Anonymization.2025​​ IEEE International Conference on​​​‌ Acoustics, Speech, and Signal‌ Processing (ICASSP 2025)Hyderabad,‌​‌ IndiaApril 2025HAL​​DOIback to text​​​‌
  • 36 inproceedingsN.Natalia‌ Tomashenko, E.Emmanuel‌​‌ Vincent and M.Marc​​ Tommasi. Exploiting Context-dependent​​​‌ Duration Features for Voice‌ Anonymization Attack Systems.‌​‌Interspeech 2025Rotterdam, Netherlands​​August 2025HALback​​​‌ to text

Conferences without‌ proceedings

  • 37 inproceedingsA.‌​‌Angélica Gutiérrez Cisneros,​​ A.Alice Foucart and​​​‌ A.Angèle Brunellière.‌ Indirect Reply Processing in‌​‌ Multilingual Conversations when Inferring​​ Speaker Meaning: an ERP​​​‌ study.ISP 2025‌ - 17th International Symposium‌​‌ of PsycholinguisticsBarcelona, Spain​​May 2025HALback​​​‌ to text
  • 38 inproceedings‌D.Dominique Knutsen,‌​‌ W. S.William S.​​ Horton and A.Angèle​​​‌ Brunellière. Faces and‌ voices in dialogue: How‌​‌ partner-specific cues contribute to​​ conversational memory.The​​​‌ Annual Meeting of the‌ Society for Text and‌​‌ DiscoursePadua, ItalyJuly​​ 2025HALback to​​​‌ text

Doctoral dissertations and‌ habilitation theses

Reports‌​‌ & preprints

Other scientific​​ publications