2024Activity reportProject-TeamTAU

7.1.1 Codalab

• Keywords:
  Benchmarking, Competition
• Functional Description:

  Challenges in machine learning and data science are competitions running over several weeks or months to resolve problems using provided datasets or simulated environments. Challenges can be thought of as crowdsourcing, benchmarking, and communication tools. They have been used for decades to test and compare competing solutions in machine learning in a fair and controlled way, to eliminate “inventor-evaluator" bias, and to stimulate the scientific community while promoting reproducible science. Current production infrastructure has been consolidated in 2021 (sovereign distributed storage, 20 GPU workers) thanks to the sponsorship of Région Ile-de-France, ANR, Université Paris-Saclay, CNRS, INRIA, and ChaLearn, to support 20,000 new users (2024), organizing or participating each year to hundreds of competitions. Some of the areas in which Codalab is used include Computer vision and medical image analysis, natural language processing, time series prediction, causality, and automatic machine learning. Codalab has been selected by the Région Ile de France to organize industry-scale challenges. Codalab has been ranked first on scientific criteria, in an independent international study: https://mlcontests.com/state-of-competitive-machine-learning-2023/. TAU continues expanding Codalab to accommodate new needs, including teaching. Link to the historical server (read-only) https://competitions.codalab.org.

  @article{codalab_competitions_JMLR, author = {Adrien Pavao and Isabelle Guyon and Anne-Catherine Letournel and Dinh-Tuan Tran and Xavier Baro and Hugo Jair Escalante and Sergio Escalera and Tyler Thomas and Zhen Xu}, title = {CodaLab Competitions: An Open Source Platform to Organize Scientific Challenges}, journal = {Journal of Machine Learning Research}, year = {2023}, volume = {24}, number = {198}, pages = {1–6}, url = {http://jmlr.org/papers/v24/21-1436.html} }

• URL:
  http://­competitions.­codalab.­org
• Contact:
  Isabelle Guyon

7.1.2 Cartolabe

• Name:
  Cartolabe
• Keyword:
  Information visualization
• Functional Description:

  The goal of Cartolabe is to build a visual map representing the scientific activity of an institution/university/domain from published articles and reports. Using the HAL Database, Cartolabe provides the user with a map of the thematics, authors, and articles. ML techniques are used for dimensionality reduction, cluster, and topic identification, visualization techniques are used for a scalable 2D representation of the results.

  Cartolabe has, in particular, been applied to the Grand Debat dataset (3M individual propositions from French Citizen, see https://cartolabe.fr/map/debat). The results were used to test both the scaling capabilities of Cartolabe and its flexibility to non-scientific and non-English corpora. We also added sub-map capabilities to display the result of a year/lab/word filtering as an online generated heatmap with only the filtered points to facilitate the exploration. Cartolabe has also been applied in 2020 to the COVID-19 Kaggle publication dataset (Cartolabe-COVID project) to explore these publications.

• URL:
  http://­www.­cartolabe.­fr/
• Publication:
  hal-02499006
• Contact:
  Philippe Caillou
• Participants:
  Philippe Caillou, Jean Daniel Fekete, Michèle Sebag, Anne-Catherine Letournel, Hande Gozukan
• Partners:
  LRI - Laboratoire de Recherche en Informatique, CNRS

7.1.3 DeepHyper

• Keywords:
  Deep learning, Autotuning, HPC
• Functional Description:
  Machine learning algorithms are continually evolving to serve diverse applications, yet their development often entails a significant trial-and-error process to identify optimal learning pipelines.This is compounded by the multitude of data preprocessing techniques, prediction (or generative) models, and learning procedures available, each offering a range of configurable parameters, also referred to as hyperparameters. DeepHyper addresses this challenge by automating the selection and configuration of algorithms and their corresponding hyperparameters, facilitating a streamlined approach for engineers and scientists to comprehend and optimize the learning pipeline. At its core, DeepHyper employs parallel Bayesian optimization, validated through rigorous testing involving up to 8,000 parallel tasks. This methodology is adaptable for both single and multi-objective tasks, enabling efficient early discarding of costly training steps. Furthermore, DeepHyper seamlessly integrates with various parallel backends, including multi-threading, multi-processing, Clouds (via the Ray library), and MPI-based schedulers on supercomputers, enhancing its scalability and versatility across different computing environments. The development of DeepHyper is supported by the TAU-team through advances in learning theory for improving and explaining its core algorithms.
• URL:
  https://­github.­com/­deephyper/­deephyper
• Contact:
  Romain Egele

7.1.4 OmniPrint

• Keyword:
  Open data
• Functional Description:

  Benchmarks and shared datasets have been fostering progress in deep learning. While there is an increasing number of available datasets, there is a need for larger ones. However, collecting and labeling data is time-consuming and expensive, and systematically varying environmental conditions is difficult and necessarily limited. Therefore, resorting to artificially generated data is helpful to drive fundamental research in deep learning. OmniPrint is geared to generating an unlimited amount of printed characters.

  Character images provide excellent benchmarks for deep learning problems because of their relative simplicity and visual nature while opening the door to high-impact real-life applications. A conjunction of technical features is required to meet our specifications: pre-rasterization manipulation of anchor points, post-rasterization distortions, natural background and seamless blending, foreground filling, anti-aliasing rendering, and importing new fonts and styles. Modern fonts such as TrueType or OpenType are made of straight line segments and quadratic Bezier curves, connecting anchor points. Thus, it is easy to modify characters by moving anchor points. This allows users to perform vectors-space pre-rasterization geometric transforms (rotation, shear, etc.) as well as distortions (e.g., modifying the length of ascenders of descenders) without incurring aberrations due to aliasing when transformations are done in pixel space (post-rasterization).

  The key technical contributions include implementing transformations and styles such as elastic distortions, natural background, foreground filling, and so on, selecting characters from the Unicode standard to form alphabets from more than 20 languages around the world, further grouped into partitions, to facilitate creating meta-learning tasks, identifying fonts, implementing character rendering with a low-level FreeType font rasterization engine, which enables direct manipulation of anchor points, adding anti-aliasing rendering, implementing and optimizing utility code to facilitate dataset formatting. To our knowledge, OmniPrint is the pioneering text image synthesizer geared toward ML research, supporting pre-rasterization transforms, which allows Omniprint to imitate handwritten characters to some degree. More details can be found in the paper (https://openreview.net/forum?id=R07XwJPmgpl, https://arxiv.org/abs/2201.06648).

• URL:
  https://­github.­com/­SunHaozhe/­OmniPrint
• Contact:
  Haozhe Sun

7.1.5 codabench

• Keywords:
  Competition, Benchmarking
• Functional Description:

  Obtaining standardized crowdsourced benchmark of computational methods is a major issue in data science communities. Dedicated frameworks enabling fair benchmarking in a unified environment are yet to be developed. Here we introduce Codabench, an open-source, community-driven platform for benchmarking algorithms or software agents versus datasets or tasks. Codabench, released in summer 2023, is the follower of Codalab, enabling the same features and more: inverted data challenges, better user experience, easier platform administration and robustness. Competition design is backward compatible allowing an easy migration from Codalab to Codabench. A public instance of Codabench (https://codabench.org) is open to everyone, free of charge, and allows benchmark organizers to compare fairly submissions, under the same setting (software, hardware, data, algorithms), with custom protocols and data formats. Codabench has unique features facilitating the organization of benchmarks flexibly, easily and reproducibly, such as the possibility of re-using templates of benchmarks, and supplying compute resources on-demand. In 2024, Codabench has registered more than 10,000 new users and computed near 80,000 participants submissions.

  @article{codabench, title = {Codabench: Flexible, easy-to-use, and reproducible meta-benchmark platform}, author = {Zhen Xu and Sergio Escalera and Adrien Pavão and Magali Richard and Wei-Wei Tu and Quanming Yao and Huan Zhao and Isabelle Guyon}, journal = {Patterns}, volume = {3}, number = {7}, pages = {100543}, year = {2022}, issn = {2666-3899}, doi = {https://doi.org/10.1016/j.patter.2022.100543}, url = {https://www.sciencedirect.com/science/article/pii/S2666389922001465} }

• URL:
  https://­www.­codabench.­org/
• Contact:
  Isabelle Guyon
• Partner:
  Région Île-de-France

7.1.6 pyriemann-qiskit

• Keywords:
  Quantum programming, Riemannian geometry, Symmetric positive definite matrices
• Functional Description:
  Literature on quantum computing suggests it may offer an advantage compared with classical computing in terms of computational time and outcomes, such as for pattern recognition or when using limited training sets. Bulding on the Qiskit library on quantum computing, pyriemann-qiskit implements a wrapper around quantum-enhanced support vector classifiers (QSVCs) and variational quantum classifiers (VQCs), to use quantum classification with Riemannian geometry. It also introduces a quantum version of the MDM algorithm, a classifier operating on the manifold of symmetric positive definite matrices.
• URL:
  https://­pyriemann-qiskit.­readthedocs.­io/­en/­latest/
• Publication:
  https://hal.science/hal-04040814v1
• Contact:
  Sylvain Chevallier
• Partner:
  IBM

7.1.7 pyriemann

• Keywords:
  Riemannian geometry, Hermitian positive definite matrices, Symmetric positive definite matrices
• Functional Description:

  Pyriemann is a Python machine learning package based on scikit-learn API. It provides a high-level interface for processing and classification of real (resp. complex)-valued multivariate data through the Riemannian geometry of symmetric (resp. Hermitian) positive definite (SPD) (resp. HPD) matrices.

  pyRiemann aims at being a generic package for multivariate data analysis but has been designed around biosignals (like EEG, MEG or EMG) manipulation applied to brain-computer interface (BCI), estimating covariance matrices from multichannel time series, and classifying them using the Riemannian geometry of SPD matrices. It is widely used in the scientific community with more than one million download.

• URL:
  https://­pyriemann.­readthedocs.­io
• Contact:
  Sylvain Chevallier

7.1.8 braindecode

• Keywords:
  Brain-Computer Interface, Deep learning
• Functional Description:
  BrainDecode is an open-source Python toolbox for decoding raw electrophysiological brain data with deep learning models. It includes dataset fetchers, data preprocessing and visualization tools, as well as implementations of several deep learning architectures and data augmentations for analysis of EEG, ECoG and MEG. It is design for neuroscientists who want to work with deep learning and deep learning researchers who want to work with neurophysiological data.
• URL:
  https://­braindecode.­org/­stable/­index.­html
• Contact:
  Sylvain Chevallier
• Partner:
  Roche

7.1.9 MOABB

• Name:
  Mother of all BCI Benchmarks
• Keywords:
  Brain-Computer Interface, Open data, Benchmarking
• Functional Description:
  Mother of all BCI Benchmarks (MOABB) allows to build a comprehensive benchmark of popular brain-computer interface algorithms applied on an extensive list of freely available EEG datasets. This is an open science initiative, serving as a reference point for the future algorithmic developments. Build on reference libraries like scikit-learn and MNE-python, machine learning pipelines can be ranked and promoted on a website, providing a clear picture of the different solutions available in the field. This software has 80k downloads and an active international development community.
• URL:
  https://­neurotechx.­github.­io/­moabb/
• Contact:
  Sylvain Chevallier

7.1.10 dnadna

• Name:
  Deep Neural Architectures for DNA
• Keywords:
  Deep learning, Population genetics
• Functional Description:
  DNADNA provides utility functions to improve development of neural networks for population genetics and is currently based on PyTorch. In particular, it already implements several neural networks that allow inferring demographic and adaptive history from genetic data. Pre-trained networks can be used directly on real/simulated genetic polymorphism data for prediction. Implemented networks can also be optimized based on user-specified training sets and/or tasks. Finally, any user can implement new architectures and tasks, while benefiting from DNADNA input/output, network optimization, and test environment.
• URL:
  https://­mlgenetics.­gitlab.­io/­dnadna/
• Contact:
  Flora Jay

8.1.1 Model frugality

In Manon Verbockhaven's PhD thesis, we study how to optimally grow a neural network architecture, to increase the performance (in terms of loss) while keeping the network as small as possible (in particular, avoiding redundancy). We showed 26 how to formulate the notion of "expressivity bottleneck" in an easily computable manner, and obtain optimal neuron weights as the result of a small SVD. We showed that the approach can scale up, with an experiment using ResNet18 on CIFAR-100. With Barbara Hajdarevic's internship, we had also started to extend the addition of neurons to existing layers to the addition of layers to an existing computation graph. Thanks to the European project MANOLO and to an ENS grant (CDSN), this work has been continued by new PhD students, Styliani Douka and Théo Rudkiewicz, and a post-doc, Stéphane Rivaud. In particular, we now allow the architecture to grow as an arbitrary DAG (Directed Acyclic Graph) 33 (paper accepted at ESANN 2025).

Within the context of Alice Lacan 's PhD (defended on Feb 4., 2025; coll. U. Evry), we applied generative modelling for data augmentation in transcriptomics 107. The high computational requirements of mainstream generative models (GAN, WGAN, diffusion model) for such high dimensional domains led to the design of a new frugal generative modelling approach, based on density alignment 34.

Within the context of Nicolas Atienza 's PhD (to be defended in March 2025; Cifre Thales, coll. GALAC@LISN), the latent space of a trained teacher is decomposed using Information Bottleneck principles. The core latent space is exploited to learn a frugal student, by distillation of the teacher.

8.1.2 Data frugality

Frugal learning is also investigated along the line of the design of data frugal algorithms. It is a major challenge in the context of domain adaptation and transfer learning, to ensure that relevant representation are learned from limited data accessibility. In the context of time serie prediction and brain signal decoding, 10 reduces the number of sensors required to obtain state-of-the-art results and 46 further explore Riemannian deep learning for domain adaptation. Riemannian models could also encompass more complex representations, using trajectory in the space of spectral frequency, to generate more robust decoding of time series 44.

Another line of research concerns the dataset alignment methods: 17 proposes a systematic exploration to avoid negative transfer effects, 41 adds data augmentation in the alignment process, 36 integrates physics-informed constraints for aligning datasets while mitigating the heterogeneity in the dimension of the data, or fighting dataset shift both in time series and labels 37.

Also, within the context of Vincenzo Schimmenti 's PhD 122, we demonstrate 25 that the information contained in GPS stations, that monitor the Earth surface deformation, can be leveraged to predict aftershocks to large earthquakes, reaching a balanced accuracy of 70%. This is true both for a very robust model (logistic regression, 2 parameters) and for our proof-of-concept CNN, which we manage to avoid to overfit using a robust ensembling technique, despite the very small number of training samples (48 spatial maps only).

8.2.1 Causal Learning

Causal learning is commonly regarded as a key research direction to enforce the properties of good AI models in terms of explainability, verifiability and fairness. Its importance is acknowledged through the PEPR-IA-Causalit-AI, starting in 2024, gathering four French laboratories/teams (Loria at Nancy; TAU and CELESTE at UPSaclay; LIG at Grenoble)

In the last year of Shuyu Dong's postdoc (partnership Fujitsu), building upon previous results 92, 93, we have tackled the notorious lack of scalability of causal graph learning from observational data. The proposed approach, called DCILP, is a Divide-and-Conquer approach. Formally, sub-problems involving the Markov blanket of each variable are defined and solved (thus with moderate complexity); the reconciliation of these partial solutions is formulated as an integer linear programming problem, reaching a very good trade-off between time-complexity and accuracy 32, recently accepted at AAAI'25.

Audrey Poinsot's PhD (Cifre Ekimetrics) is concerned with counter-factual reasoning in the context of strategic and marketing decisions. Because data in that area does not pertain to Big Data, the PhD first focused on Data Augmentation in Causal context 118, as known causality links can be leveraged to ease learning. But the lack of recognized benchmarks in that area led us to propose a comprehensive survey of deep structural causal models, focusing on their ability to answer counterfactual queries using observational data within known causal structures 40.

Armand Lacombe's PhD, defended on March 5th, 2024 52, tackles the identification of the conditional average treatment effet in the case where the control and treated distributions are different (which is the usual case in practice). A thorough theoretical and empirical analysis have been proposed for the original approach, relying on the design of two latent representations.

A collaboration with Inria Nerv team, from Paris-Brain Institute, investigates the causal chains found in the brain activity. Motor imagery and preparation of movement executions is build on sequential activations of neural assemblies within the brain. In 13, an avalanche-based approach is designed to reconstruct those causal chain of activations in brain imaging.

Causality is also at the core of TAU participation in the INRIA Challenge OceanIA, that started in 2021 121. The main challenge is related to out-of-distribution learning, motivated by the analysis of the TARA images to identify the ecosystems in the diverse sites of the data collection. The high imbalance of the data among the classes, the prevalence of outliers, suggest that the use of multi-modal embeddings as explored in N. Atienza's PhD 27 (see below) might support the design of relevant metrics in the considered space. A post-doc has been hired, who will start in February 2025.

8.2.2 Explainable Learning

Nicolas Atienza's PhD (Cifre Thales, co-supervised with Johanne Cohen, LISN; 2 patents Thales pending) is to tackle the main three goals of Trustable AI, i.e. explainability, reliability and frugality. Toward explainability, a new approach to build conceptual model explanations, called CB2 (Cut the Black Box) proceeds by combining multi-modal embeddings and Multi-Criteria Decision Aid 27. Toward reliability, the new Sample-efficient Probabilistic Detection using Extreme Value Theory (SPADE) transforms a classifier into an abstaining classifier, offering provable protection against out-of-distribution and adversarial samples.

During the last year of the Horizon Europe project TRUST-AI, and even though our work within this project was completed, we extended the results obtained on Memetic Semantic Genetic Programming (MSGP) 111. MSGP is able to generate short, and hence hopefully easily explainable expressions for Symbolic Regression problems. We implemented a boosting procedure around MSGP, which improved the performances without degrading too much the interpretability 19.

Also, note that the work described in Section 8.4.1 about Interpretable Learning Effective Dynamics (iLED) framework 21 also contributes to this line of research, adding interpretability to the Deep Learning approach to dynamical systems simulations.

8.2.3 Improved Learning

The research in AutoML is gently fading out at TAU, with Isabelle Guyon's departure for Google Brain and Michèle Sebag's and Marc Schoenauer's retirements (even if emeritus, they cannot supervise new PhD students nor can they be PIs for new projects). On the other hand, the long-lasting expertise of the team in terms of black-box optimization proved to be useful in order to globaly improve the learning process, a priori or post-hoc.

Romain Egele's PhD (coll Argonne National Labs, USA), defended in June 2024 51 was focused on Hyper-Parameter Optimization (HPO) and Neural Architecture Search (NAS), and the deployment of AI and HPC. The last contrbution in his thesis 60 complements his previous work in developing DeepHyper, a package allowing users to conduct NAS with genetic algorithms using TensorFlow or PyTorch 99. A benchmark of early discarding strategies was conducted to compare state of the art algorithms. It was noticed that a very simple strategy, dubbed 1-Epoch, performed significantly better when “computing duration” is the bottleneck. A method based on Bayesian regression (including both aleatoric and epistemic uncertainties) for learning curve extrapolation was also proposed and dubbed Robust Bayesian Early Rejection (RoBER) 98; Best paper Award at IEEE International Conference on e-Science.

Mathurin Videau is a CIFRE PhD student with Meta (that started long before Meta had to obey Musk's worst practices), co-supervised by Olivier Teytaud (former member of the team before joining Meta). Mathurin's PhD focuses on post-hoc improvements of fully trained models by using black-box algorithms to optimize an impactful but small part of model. Using BBO allows us to optimize non-differentiable loss functions that match the user's true goals more efficiently than the loss used for the initial training of the model by standard backpropagation, loss that needs to be differentiable 71 (submitted). This allows for instance to exactly optimize the Word Error Rate in translation tasks, the number of deaths in Doom RL agents, or to let the user interactively guide generative processes by directly acting in the latent space 73 (submitted).

8.2.4 Towards high-quality and private genomes based on generative neural networks

In collaboration with the Institute of Genomics of Tartu, we have been leveraging two types of generative neural networks (Generative Adversarial Networks and Restricted Boltzmann Machines) to learn the high dimensional distributions of real genomic datasets and create artificial genomes 129. These artificial genomes retain important characteristics of the real genomes (genetic allele frequencies and linkage, hidden population structure, ...) without copying them and have the potential to be valuable assets in future genetic studies by providing anonymous substitutes for private databases (such as the ones hold by companies or public institutes like the Institute of Genomics of Tartu).

The main challenges lie in scaling up to the full genome and in making sure that no personal genetic data is leaked. For this, we had developed various deep learning generative architectures, from plain GANs and RBMs, to convolutional GANs, with or without attention. Following this body of work where models are trained in the SNP data space (ie., the space of DNA sequences, removing sites that are constant accross the dataset), we propose in 45 a conceptually different approach. Our method combines dimensionality reduction, achieved by Principal Component Analysis (PCA), and a Generative Adversarial Network (GAN) learning in this reduced space, which is much smaller when facing datasets with fewer individuals (5 000) than SNP sequence length (60 000). Such a low ratio between number of samples and sample dimension makes the task proned to overfitting and calls for careful check of possible privacy leaks. We studied various privacy scores, including in particular the AATS metric (nearest neighbor adversarial accuracy) proposed by 128, and sorted the different models according to quality and privacy scores.

Furthermore, we proposed an alternative approach based on a diffusion model, which had never been investigated in the context of population genetics 68. We tested how these synthetic genomes could replace or augment reference databases for local ancestry inference (LAI). LAI is a major task in population genetics which consists in identifying the origin of genomic regions along the chromosomes of a given indiviual. It relies on reference panels that should be as diverse as possible to avoid biaises and could thus highly benefit from accurate data augmentation, in particular for underrepresented populations.

A next challenging step is to design interpretable generative models capable of handling genotypes and phenotypes jointly. We have indeed recently shown, in a supervised setting, that classical neural networks combined with post-hoc interpretation techniques yielded insights on the relationships linking genetic loci and phenotypes 72. In parallel, we aim to investigate the potential of generative modeling for this task (collaboration with U Tartu).

8.4.1 ML and Reduced Order Models for Dynamical Systems

Participants: Michele Alessandro Bucci, Marc Schoenauer, Emmanuel Menier, Thibault Monsel, Mouadh Yagoubi (IRT-SystemX), Lionel Mathelin (DATAFLOT team, LISN), Onofrio Semeraro (DATAFLOT team, LISN), Petros Koumoutsakos (Harvard SEAS), Sebastian Kaltenbach (Harvard SEAS).

Numerical simulations of fluid dynamics in industrial applications require the spatial discretization of complex 3D geometries with consequent demanding computational operations for the PDE integration. The computational cost is mitigated by the formulation of Reduced Order Models (ROMs) aiming at describing the flow dynamics in a low dimensional feature space. The Galerkin projection of the driving equations onto a meaningful orthonormal basis speeds up the numerical simulations but introduces numerical errors linked to the underrepresentation of dissipative mechanisms.

Emmanuel Menier's PhD, defended in January 2024 54 trained a DNN to compensate missing information in the projection basis. By exploiting the projection operation, the ROM correction consists in a forcing term in the reduced dynamical system which has to: i) recover the information living in the subspace orthonormal to the projection one; ii) ensure that its dynamics is dissipative. A constrained optimization is then employed to minimize the ROM errors but also to ensure the reconstruction and the dissipative nature of the forcing improving the prediction while preserving the guarantees of the ROM. The approach was extended on Michelin use case of rubber calendering process.

During his PhD thesis, Emmanuel Menier also spent 3 months in Spring 2023 in Prof. Petros Koumoutsakos' group at SEAS - Harvard, John A. Paulson School of Engineering and Applied Sciences. It was a perfect match between his previous work and the group's expertise in high dimensional dynamical complex system (e.g., CFD), and resulted in the Interpretable Learning Effective Dynamics (iLED) framework, a novel framework based on nonlinear dimension reduction thanks to deep neural networks, that offers comparable accuracy to state-of-the-art recurrent neural network-based approaches while providing the added benefit of interpretability 21. The basic idea of iLED is grounded on the Mori-Zwanzig formalism, an approach that has been later generalized to other dynamical systems 116.

Thibault Monsel's PhD has been indeed focusing on the learning of dynamical sytems involving delays, i.e. Delayed Differential Equations (DDE). While Neural ODE was a conceptual breakthrough, it cannot learn partially-observable dynamical systems. Inspired by the Mori-Zwanzig formalism and Takens theorem, we develop another way to extract this information, using delays, i.e. using past observable states of the system 116, 64. These delays may depend on the current state. Thibault contributed to implement efficiently Delayed Differential Equations in deep learning frameworks (Ajax, PyTorch) and showed the advantage of DDEs over (Augmented) Neural DDEs and recurrent networks (LSTM) under certain circonstances, in particular in the case of constant delays, that can now be learned during training 38.

8.4.2 Graph Neural Networks for Numerical Simulations

Participants: Guillaume Charpiat, Michele Alessandro Bucci (Safran Tech), Marc Schoenauer, Matthieu Nastorg, Lionel Mathelin (DATAFLOT team, LISN), Thilbault Faney (IFPEN), Jean-Marc Gratien (IFPEN).

During the 2.5 years that he spent at TAU (2021-2023), Alessandro MIchele Bucci, now at Safran Tech, worked on several use cases of IFPEN, with the goal of accelerating some softwares that IFPEN uses daily. This IFPEN/TAU collaboration led to a successful DataIA proposal, ML4CFD, that funded Matthieu Nastorg's PhD, defended in March 2024 55. After making significant improvements on B. Donon’s Deep Statistical Solvers (DSS) 97, replacing the arbitrary number of iterations of the message passing mechanism by the solution of a fixed point equation 22, the final part of his PhD by considered the DSS approach as a preconditionner for a domain decomposition method 39.

Alessandro also contributed to propose a new approach to maintain the physical consistancy of GNN based approach to data assimilation for the RANS (Reynolds-Averaged Navier Stokes) equations, by hybridization with the classical adjoint metho 66.

Last but not least, we successfull applied to an Action Exploratoire (PI Guillaume Charpiat) Large Physics Models to investigate the use of Transformers and Large Foundational Models for Numerical simulations. Matthieu Nastorg has been hired on this AeX and he will co-supervise a new PhD student. This AeX is also tightly linked with the startup company AUGUR, founded by Emmanuel Menier, Matthieu Nastorg and Alice Lacan, all former PhD students in the team.

8.4.3 Advances in sparse recovery for inverse problem and application in M/EEG

Participants: Matthieu Kowalski, Jean-Baptiste Malagnoux, Diego Delle Donne (Essec), Leo Liberti (LiX), Benoît Malézieux (Inria Mind), Pierre Barbault (CentraleSupelec), Thomas Moreau (Inria Mind), Charles Soussen (CentraleSupelec).

Inverse problems involve reconstructing underlying signals or images from indirect or incomplete measurements and often require additional constraints or regularization to ensure unique and stable solutions. Sparse coding addresses these challenges by representing signals with a small number of nonzero coefficients in a suitable basis or dictionary. Methods like Convolutional Dictionary Learning build on this principle and have been successfully applied in areas such as neuroimaging and audio signal analysis.

In 35, we have established a theoretical equivalence between Convolutional Dictionary Learning (CDL) and Non-Negative Matrix Factorization (NMF) methods for signal processing in the time-frequency domain. We show that signals represented using CDL, which relies on sparse coding, can also be synthesized using factorized time-frequency coefficients in semi-NMF or complex-NMF forms. This connection bridges two widely-used approaches, highlighting their potential for joint optimization in applications like music transcription and biomedical signal analysis.

In 15 we provide groundbreaking perspective on the $L_2$+$L_0$ sparse approximation problems, introducing a comprehensive, big-M independent integer linear programming formulation. This development effectively circumvents the shortcomings of existing methods, ensuring the accurate recovery of global minimizers without requiring predefined bounds.

58 introduces LEMUR, an Expectation-Maximization (EM) framework for estimating Bernoulli-Gaussian model parameters in inverse problems. By jointly estimating both the signal and hyperparameters, LEMUR reduces the need for manual tuning and proves effective for structured inverse problems, particularly with correlated measurement operators. Complementing this practical approach, 57 provides a theoretical study on Maximum A Posteriori (MAP) estimation for sparse recovery using the Bernoulli-Gaussian model. It establishes connections between joint-MAP and marginal-MAP estimators and classical methods like the Lasso and Sparse Bayesian Learning, demonstrating how specific relaxations of the MAP problem can efficiently recover the support of sparse signals.

63 explores the feasibility and limitations of prior learning in unsupervised inverse problems, particularly focusing on dictionary learning and structured priors. The study shows that recovering a dictionary from incomplete data is only possible when multiple measurement operators span the entire signal space or when weak priors, such as equivariance or group structures, are incorporated. It highlights that handcrafted priors can be outperformed by learned priors, but only when they are well-adapted to the specific inverse problem. Complementing this theoretical perspective, 62 introduces a practical approach to prior learning, leveraging structured optimization techniques to refine sparse signal representations. This work provides experimental validation of learned priors in various inverse problem settings, demonstrating their potential for improving reconstruction quality compared to traditional handcrafted priors.

8.4.4 Instrument recognition in MIR

Participants: Matthieu Kowalski, Dylan Sechet, Francesca Bugiotti (LISN), Edouard d'Hérouville (Linkaband), Filip Langiewicz (Linkaband).

Detecting musical instruments in audio recordings is a complex challenge in Music Information Retrieval (MIR), particularly in polyphonic settings where multiple instruments play simultaneously. While existing methods perform well for common instruments with abundant training data, they often fail to detect rare or underrepresented instruments, such as those found in orchestral, non-Western, or niche musical contexts. This issue is further compounded by the lack of labeled datasets for rare instruments, limiting the ability of models to generalize across diverse instruments and styles.

In 43 we have proposed a hierarchical deep learning approach to address the challenge of detecting minority instruments in polyphonic music recordings. By leveraging the MedleyDB dataset, the study introduces a hierarchical classification framework that integrates group-level and instrument-level predictions to enhance the detection of rare instruments. The proposed two-pass classification approach demonstrates improved performance for underrepresented instruments.

8.4.5 Simulations for evolutionary genomics

Participants: Guillaume Charpiat, Flora Jay, Léo Planche, Arnaud Quelin.

Collaboration: Bioinfo Team (LISN), MNHN (Paris), UNAM (Mexico), U Brown (USA), METU (Turkey).

In population genetics, simulators are a valuable resource, enabling the testing of tool robustness, comparing the outcomes of stochastic evolutionary models with real observations, and performing simulation-based inference. In the latter case, simulations allow us to work in a supervised setting to solve the inverse problem of inferring the simulator’s evolutionary parameter inputs from genomic outputs. We previously demonstrated how machine learning and deep learning could contribute to this task. More recently, we have focused on using simulations to test and infer individuals' relatedness and population structure from ancient DNA 18, 115, two key questions frequently asked in paleogenomics, such as in the study of Neolithic population dynamics in Anatolia 61 (collaboration with METU, Turkey).

8.6.1 Labor Studies

Participants: Philippe Caillou, Michèle Sebag, Guillaume Bied, Armand Lacombe, Solal Nathan, Hande Gozukan, Jean-Pierre Nadal (EHESS), Bruno Crépon (ENSAE).

Job markets The DataIA project Vadore 78 (partners ENSAE and Pôle Emploi/France Travail) benefits from the sustained cooperation and from the wealth of data gathered by France Travail. The data management is regulated along a 3-partite convention (GENES-ENSAE, Univ Paris-Saclay, Pôle Emploi). Extensive efforts have been required to achieve the data pipelines required to enable learning recommendation models and exploiting them in a confidentiality preserving way (G. Bied's PhD, 50). The acceptability of the algorithm for the job seekers has been investigated using large-scale (100,000 job seekers) in Feb. 2023 and June 2024.

An important criterion, besides the performance in terms of recall and the time-to-solution, regards the fairness of the recommendation model 77, 76. A comprehensive study examining gender-related gap in several utilities (wages, types of contract, distance-to-job) has been conducted, comparing the gaps observed in actual hirings, in applications, and in recommendations. Interestingly, the gap in recommendations closely mimics that in actual hirings and in applications (if any, the recommendation algorithm tends to decrease the gap). Algorithmic fairness in domains as sensitive as employment is under scrutiny of French and European regulations. The difficulty is to decouple the biases observed in applications (thus reflecting job seekers' preferences, that should be respected) from those due to recruiters (that should not be perpetuated in the learned models).

Another criterion concerns the congestion of the job market (share of job offers paid attention to by job seekers). Recommender systems tend to increase the congestion due to the so-called popularity bias. Early attempts to prevent the congestion have been investigated in 79, using optimal transport.

Both fairness and congestion issues are at the core of S. Nathan's PhD (coll. Univ. Ghent, Belgium). A first research direction is concerned with integrating the congestion in the recommendation loss; this requires a global view of the market dynamics, and the difficulty is to design a loss term that is both computationally affordable and differentiable. A second research direction along this line is to integrate an estimation of job offer popularity within the recommendation system, enabling job seekers to anticipate and react to, competition. Interestingly, such compound architecture (integrating the job offer popularity estimate and its effects on the decision of applying) could enable to model competition-avoidance strategies, in particular in relation with gender effects.

A key difficulty for research on ML-based job recommendation is the lack of open and representative datasets, owing to the very sensitive nature of the data and the protection of vulnerable persons. We collaborate with U. Gent, Belgium, welcoming Guillaume Bied for his post-doc and Solal Nathan for a PhD visit, on this topic.

8.6.2 Health and practices

Participants: Philippe Caillou, Michèle Sebag, Armand Lacombe, Cyriaque Rousselot, Olivier Allais (INRA), Julia Mink (Univ. Bonn, DE), Florian Yger (INSA Rouen).

Continuing our former partnership with INRAE (in the context of the Initiative de Recherche Stratégique Nutriperso; 101), we proposed the HorapestDataIA project to uncover the potential causal relationships between pesticide dissemination and children's health (Cyriaque Rousselot's PhD). Medical data is accessed using the Health data Hub after the CNIL approval and proprietary data from INRAE is used for detailed pesticide purchase on every french parcel. Contacts have been taken with the CHU Toulouse for cooperation on complementary data.

8.6.3 Scientific Information System and Visual Querying

Participants: Philippe Caillou, Michèle Sebag, Anne-Catherine Letournel, Hande Gozukan, Jean-Daniel Fekete (AVIZ, Inria Saclay).

A third area of activity concerns the 2D visualisation and querying of a corpus of documents. Its initial motivation was related to scientific organisms, institutes or Universities, using their scientific production (set of articles, authors, title, abstract) as corpus. The Cartolabe project (see also Section 7) started as an Inria ADT (coll. Tao and AVIZ, 2015-2017). It received a grant from CNRS (coll. Tau, AVIZ and HCC-LRI, 2018-2019).

The originality of the approach is to rely on the content of the documents (as opposed to, e.g. the graph of co-authoring and citations). This specificity allowed to extend Cartolabe to various corpora, such as Wikipedia, Bibliotheque Nationale de France, or the Software Heritage. Cartolabe was also applied in 2019 to the Grand Debat dataset: to support the interactive exploration of the 3 million propositions; and to check the consistency of the official results of the Grand Debat with the data. Cartolabe has also been applied in 2020 to the COVID-19 kaggle publication dataset (Cartolabe-COVID project) to explore these publications.

Among its intended functionalities are: the visual assessment of a domain and its structuration (who is expert in a scientific domain, how related are the domains); the coverage of an institute expertise relatively to the general expertise; the evolution of domains along time (identification of rising topics). A round of interviews with beta-user scientists has been performed in 2019-2020. Cartolabe usage raises questions at the crossroad of human-centered computing, data visualization and machine learning: i) how to deal with stressed items (the 2D projection of the item similarities poorly reflects their similarities in the high dimensional document space; ii) how to customize the similarity and exploit the users' feedback about relevant neighborhoods. A statement of the current state of the project was published in 2021 83.

10.1.1 Visits to international teams

10.2.1 Horizon Europe

10.2.2 H2020 projects

10.3.1 ANR

• Chaire IA HUMANIA 2020-2024 (600kEuros), Democratizing Artificial Intelligence.

  Coordinator: Isabelle Guyon (TAU)

  Participants: Marc Schoenauer, Michèle Sebag, Anne-Catherine Letournel, François Landes.

• PEPR IA SAIF (400k€) Safe AI through Formal methods

  Coordinator: Caterina Urban (INRIA Antique)

  Participant: Guillaume Charpiat.

• PEPR IA CAUSALI-T-AI (400k€) CAUSALIty Teams up with Artificial Intelligence

  Coordinator: Marianne Clausel (Université de Loraine)

  Participant: Michèle Sebag, Alessandro Leite.

• RoDAPoG 2021-2025 (302k€) Robust Deep learning for Artificial genomics and Population Genetics

  Coordinator: Flora Jay,

  Participants: Cyril Furtlehner, Guillaume Charpiat.

• SPEED 2021-2024 (49k€) Simulating Physical PDEs Efficiently with Deep Learning

  Coordinator: Lionel Mathelin (LISN (ex-LIMSI))

  Participants: Michele Alessandro Bucci, Guillaume Charpiat, Marc Schoenauer.

10.3.2 Others

• Inria Challenge OceanAI 2021-2024, AI, Data, Models for a Blue Economy

  Coordinator: Nayat Sanchez Pi (Inria Chile)

  Participants: Marc Schoenauer, Michèle Sebag and Shiyang

• DATAIA YARN 2022-2025, Automatic Processing of Messy Brain Data with Robust Methods and Transfer Learning

  Coordinator: Sylvain Chevallier

  Participants: Florent Bouchard (L2S), Fredéric Pascal (L2S), Alexandre Gramfort (Meta), Sara Sedlar

• Fair Universe 2022-2025, We received with Lawrence Berkeley Labs a grant of 6.4 million USD to develop benchmarks in High Energy Physics and implement them on Codabench. Colaboration with David Rousseau of IJCLAB.

  Coordinator: Isabelle Guyon

  Participants: David Rousseau, Ragansu Chakkappai, Ihsan Ullah

• Action Exploratoire 2024-2026, Large Physics Models

  Coordinator: Guillaume Charpiat

  Participants: Mac Schoenauer, Matthieu Nastorg (post-doc), Theofanis Ifaistos (PhD student)

11.1.1 Scientific events: organisation

11.1.2 Scientific events: selection

11.1.3 Journal

11.1.4 Invited talks

• Guillaume Charpiat An example of AI4Science: ML4CFD at InPEx workshop (NumPEx), Sitges (Spain) June 18th 2024
• Guillaume Charpiat ML4CFD: Deep learning for numerical simulations at Dassault Systèmes & Inria Scientific Day, November 29th 2024
• Cyril Furtlehner Online feature learning in terms of spectral flow processes workshop "Complex systems, statistical mechanics and machine learning crossover (in memory of Giovanni Paladin)" in Les Houches (March 24-29 2024)
• Cyril Furtlehner Bypassing first order phase transition for RBM training workshop "From Machine-Learning Theory to Driven Complex Systems and back" Lausanne May 22-24 2024
• François Landes Learning representations of glassy liquids with roto-translation equivariant Graph Neural Networks workshop "Complex systems, statistical mechanics and machine learning crossover (in memory of Giovanni Paladin)" in Les Houches (March 24-29 2024)
• Marc Schoenauer, Intelligence artificielle générale : utopie ou dystopie ?, séminaire de philosophie "Les formes de l'intelligence" de l'université Paris Est Créteil;
• Sylvain Chevallier, Riemannian geometry applied to time series, UMR MIA, AgroParisTech, lab seminar, 19/12/2024.
• Sylvain Chevallier Open science tools for reproducibility, Open science week, Paris-Saclay, 5/11/2024
• Michèle Sebag: Machine Learning and Interventions, Harvest Alliance, AgroParisTech, 3/2/2025
• Michèle Sebag: Job Recommendation for all: Challenges, Biases and Next, From physics to neurosciences and social sciences, Conférence en l'honneur de J. P. Nadal, Ulm, 29/9/2024
• Michèle Sebag: Causal Discovery : A Divide and Conquer approach using Integer Linear Programming, Data conference, Isite; https://cap2025.fr/developpement-instrumental), 20/6/2024
• Michèle Sebag, Causal models are generative models. What is special about them ?, séminaire Collège de France, cours S. Mallat, 14/2/2024
• Michèle Sebag,IA: 10 minutes congrès de la SIF, 6-7/6/24.

11.1.5 Leadership within the scientific community

• Sylvain Chevallier: President of the academic society CORTICO, promoting the research in brain-computer interface; Executive Committee, Institut de Convergence DataIA; Research Committee, IA Cluster Paris-Saclay; Head of MSCA-Horizon Europe Cofund DeMythif.AI
• Flora Jay: member of GDR BiM science board (23-now)
• Marc Schoenauer: Advisory Board, ACM-SIGEVO, Special Interest Group on Evolutionary Computation; Chair of Advisory Board (Founding President 2015-2022), SPECIES, Society for the Promotion of Evolutionary Computation In Europe and Surroundings, that organizes the yearly series of conferences EvoStar. Senior Reviewer, IJCAI.
• Michèle Sebag: Member of scientific council of the AMIES Labex; Area Chair NeurIPS, ECML-PKDD; Senior Meta-Reviewer ECAI, nommée membre d'honneur de la SIF.

11.1.6 Scientific expertise

• Guillaume Charpiat : MdC hiring committees at DI ENS and at LIPN, USPN (Université Sorbonne Paris Nord, Villetaneuse)
• Guillaume Charpiat: Jean Zay (GENCI/IDRIS) committee member for resource allocation (GPU) demand expertise
• Guillaume Charpiat: expertise for post-doctoral grant allocation at MIAI (Institut Interdisciplinaire en Intelligence Artificielle de Grenoble)
• Sylvain Chevallier, "Conseil Scientifique", Inclusive Brain
• Sylvain Chevallier, PR hiring committees: LS2N, Nantes 19/04/2024; IUT Orsay, 26/04/2024
• Cyril Furtlehner, Inria representative at the Dataia COMP
• Marc Schoenauer, Scientific Advisory Board, BCAM, Bilbao, Spain
• Marc Schoenauer, "Conseil Scientifique", IFPEN
• Marc Schoenauer, "Conseil Scientifique", Mines Paritech
• Marc Schoenauer, "Conseil Scientifique", ADEME
• Marc Schoenauer, scientific coordinator of the IRT SystemX IA2 program (Artificial Intelligence for Augmented Engineering)
• Michele Sebag, "Conseil scientifique", IRSN
• Michele Sebag, FNRS (PhDs and Post-docs)
• Michele Sebag, hiring committee, Univ. Grenoble

11.1.7 Research administration

• Guillaume Charpiat: head of the Data Science department at LISN, Université Paris-Saclay.
• Michele Sebag, Member of Council: Institut Pascal, IRSN, ISC-PIF.
• Sylvain Chevallier, co-chair of the Scientific Council of Computer Science dept. from Universite Paris-Saclay; elected member of executive committee of University Institute

11.2.1 Teaching

• Licence : Philippe Caillou, Computer Science for students in Accounting and Management, 192h, L1, IUT Sceaux, Univ. Paris Sud.
• Licence : François Landes, Introduction to Statistical Learning, 51h, L3, Univ. Paris-Sud.
• Licence : Matthieu Kowalski, Signal Processing, L3, 25h, Univ. Paris-Saclay
• Master : Guillaume Charpiat, Deep Learning in Practice, 24h, M2 Recherche, MVA / Centrale-Supelec / DSBA / MscIA
• Master : Guillaume Charpiat, Information Theory, 14h, M1 IA Paris-Sud.
• Master: Sylvain Chevallier, Machine learning algorithms, 12h, M1, Univ. Paris-Saclay.
• Master : Isabelle Guyon, Project: Creation of mini-challenges, M2, Univ. Paris-Sud.
• Master : Flora Jay, Population genetics inference, 4h, M2, U PSaclay.
• Master: Matthieu Kowalski, Signal Processing, 25h, M2, Univ. Paris-Saclay
• Master: Matthieu Kowalski, Sparse Coding, 36h, M2, Univ. Paris-Saclay
• Master : François Landes, Foundational Principles of Machine Learning, 25h, M1 Recherche (AI track), U. Paris-Sud.
• Master : François Landes, Machine Learning, 42h, M2 Recherche, Univ. Paris-sud, physics department (PCS international Master)
• Master : Michèle Sebag, Deep Learning, 4h; Reinforcement Learning, 12h; M2 Recherche, U. Paris-Sud.
• Master : Beatriz Seoane, Applied Statistics, 25h, M1 Recherche (AI track), U. Paris-Saclay.
• Tutorial for PhDs (and others) : Guillaume Charpiat, Deep Learning for Physics, 3h
• Continuing education (ie teaching in companies): Guillaume Charpiat, Machine Learning and Deep Learning, 6 days.

11.2.2 Supervision

• PhD Emmanuel MENIER, Deep Learning for Reduced Order Modeling, from 1/9/2020, Michele Alessandro Bucci, Marc Schoenauer, and Mouadh Yagoubi (IT SystemX), defended 25/01/2024.
• PhD Armand LACOMBE, Changes of representation for counter-factual inference, Michele Sebag and Philippe Caillou, defended 05/03/2024.
• PhD Romain EGELE, Optimization of Learning Workflows at Large Scale on High-Performance Computing Systems, Isabelle Guyon/Michèle Sebag and Prasanna Balaprakash (Argonne), defended 17/06/2024.
• PhD Mathieu NASTORG, Scalable GNN Strategies to Solve Poisson Pressure Problems in CFD Simulations, Guillaume Charpiat, Marc Schoenauer and Michele Alessandro Bucci, defended 15/04/2024.
• PhD Guillaume BIED, Concevoir et évaluer les algorithmes de recommandation pour le marché du travail, 1/10/2019, Bruno Crepon (CREST-ENSAE) and Philippe Caillou, defended 10/07/2024.
• PhD Appoline MELLOT, Machine learning and domain adaptation for enhancing the measure of brain health with MEG and EEG signals, Alexandre Gramfort (Inria MIND) and Sylvain Chevallier, defended 08/11/2024.
• PhD Maria Sayu YAMAMOTO, Addressing the Large Variability of EEG Data with Riemannian Geometry: Toward Designing Reliable Brain-Computer Interfaces , Fabien Lotte (Inria Potioc) and Sylvain Chevallier, defended 02/12/2024.
• PhD Emmanuel GOUTIERRE, Machine learning-based particle accelerator modeling, Johanne Cohen (LISN/Galac) and Michèle Sebag, defended 19/12/2024
• PhD Francisco PEZZICOLI Statistical Physics - Machine Learning Interplay: from Addressing Class Imbalance with Replica Theory to Predicting Dynamical Heterogeneities with SE(3)-equivariant Graph Neural Networks, François Landes and Guillaume Charpiat, defended 19/12/2024.
• PhD Thibault MONSEL, Active Deep Learning for Complex Physical Systems, Alexandre Allauzen (LAMSADE), Guillaume Charpiat, Lionel Mathelin (LISN), Onofrio Semeraro (LISN), defended 20/12/2024.
• PhD in progress - Anaclara ALVEZ Scale-Equivariant Neural Networks from 1/11/2023, François Landes and Cyril Furtlehner.
• PhD in progress - Bruno ARISTIMUNHA PINTO deep learning for decoding electroencephalography from 01/06/2023, Raphael Y de Camargo (UFABC Brazil), Marie-Constance Corsi (Inria Nerv), Sylvain Chevallier
• PhD in progress - Nicolas ATIENZA, Towards Reliable ML: Leveraging Multi-Modal Representations, Information Bottleneck and Extreme Value Theory, from 1/4/21, Michèle Sebag and Johanne Cohen.
• PhD in progress - Nicolas BÉREUX interpretability and pattern extraction in Restricted Boltzmann Machines from 1/11/2023, Beatriz Seoane Bartolome, Cyril Furtlehner.
• PhD in progress - Eva BOGUSLAWSKI Congestion handling on Power Grid governed by complex automata, from 1/05/22, Alessandro Leite, Mathieu Dussartre (RTE) and Marc Schoenauer
• PhD in progress - Thibault DE SURREL Learning context invariant representations for EEG data, from 1/11/2023, Florian Yger (ENSICaen), Fabien Lotte (Inria Potioc), Sylvain Chevallier
• PhD in progress - Styliani DOUKA Growth strategies for neural architectures from 01/01/2024, Guillaume Charpiat and Sylvain Chevallier
• PhD in progress - Badr Youbi IDRISSI Learning an invariant representation through continuously evolving data, from 01/10/22, David Lopez-Paz (Meta) and Michèle Sebag
• PhD in progress - Jean-Baptiste MALAGNOUX Convolutional Dictionary Learning and time-frequency Nonmegative Matrix Factorization, from 1/10/2022, Matthieu Kowalski
• PhD in progress - Florent MICHEL Deep Learning for Dictionary Learning, from 1/10/2022, Matthieu Kowalski and Thomas Moreau (Inria Mind)
• PhD in progress - Solal NATHAN, Job recommendation, AI Ethics and Optimal Transport., 1/1/2023, Michèle Sebag and Philippe Caillou.
• PhD in progress - Audrey POINSOT, Causal Uncertainty Quantification under Partial Knowledge and Low Data Regimes, from 1/03/22, Nicolas Chesneau (Ekimetrics), Guillaume Charpiat, Alessandro Leite, and Marc Schoenauer
• PhD in progress - Arnaud QUELIN, Infering Human population history with approximated Bayesian computation and machine learning, from ancient and recent genomes' polymorphism data, from 1/10/22, Frédéric Austerlitz (MNHN), Flora Jay
• PhD in progress - Cyriaque ROUSSELOT, Spatio-temporal causal discovery – Application to modeling pesticides impact, from 1/10/22, Philippe Caillou
• PhD in progress - Théo RUDKIEWICZGrowing neural networks for frugal learning from 01/10/2024, Guillaume Charpiat and Sylvain Chevallier
• PhD in progress: Nilo SCHWENKE Modélisation des batteries Lithium-Ion par Physics-Informed Neural Networks from 1/09/2023, Cyril Furtlehner
• PhD in progress - Antoine SZATKOWNIK, Deep learning for population genetics, from 1/10/22, Flora Jay, Burak Yelmen, Cyril Furtlehner and Guillaume Charpiat
• PhD in progress - Manon VERBOCKHAVEN, Strategies for Neural Architecture Growth, from 11/2021, Sylvain Chevallier and Guillaume Charpiat
• PhD in progress - Sebastien VELUT Understanding and addressing within-user variability in reactive and passive Brain-Computer Interfaces since 13/11/2023, Frédéric Dehais (ISAE SupAero), Marie-Constance Corsi (Inria Nerv), Sylvain Chevallier
• PhD in progress, Mathurin VIDEAU, Reinforcement Learning with sparse reward, from 01/10/2021, Alessandro Leite, Marc Schoenauer and Olivier Teytaud (Meta).

11.2.3 Juries

• Flora Jay, PhD jury member: Guillaume Lan-Fong 13/12/24 , Aurélien Beaude 6/12/24, Letizia Lamperti 22/11/24, Luca Nesterenko 18/11/24, Margaux Lefebvre 16/09/24
• Marc Schoenauer, PhD jury member: Thimotée Anne, LARSEN team at Inria Nancy, 6/6/24; Claire Bizon Monroc, DYOGENE team, Inria Paris, 12/11/24; Michele Quattromini, LISN, 13/12/24; Abdelkader Dib, IFPEN, 14/1/25;
• Sylvain Chevallier, PhD reviewer: Mohammad Javad DARVISHI BAYAZI, MILA Montréal, 29/10/2024; Mathieu SERAPHIM, ENSICaen, 11/12/2024 Alexandre BLEUZE, GIPSA-lab, Grenoble, 8/12/2023, PhD jury president: Maxime TOQUEBIAU, ISIR, Paris, 16/12/2024; Ahmad CHAMMA, Inria Mind, 14/06/2024; Juan Jesús TORRE TRESOLS, ISAE SupAero, 21/10/2024, PhD jury member: Igor CARRARA, Inria Cronos, 12/10/2024; Fernando GONZALEZ, Coria, Rouen 29/03/2023; Armita KHAJEH NASSIRI, Team LaHDAK, LISN, 13/07/2023
• François Landes: head of the M1 and M2 AI track selection comittee (M1 and M2 combined: 1000+ applicants per year). Also head of the scholarship short-listing comittee.
• Matthieu Kowalski, PhD reviewer
• Cyril Furtlehner, PhD reviewer: Maciej KARCZ, CEA Cadarache 1/10/2024
• Michele Sebag, PhD reviewer: F. Jourdan, U. Toulouse; Simo Alami (X); Yoosof Mashayekhi, U. Gent, Belgium; Lei Zan, U. Grenoble; Colin Troisemaine, IMT Altlantique.
• Guillaume Charpiat, PhD jury member: Alexandre Vérine (Lamsade Dauphine PSL) 01/07/2024, Daniel Zyss (École des Mines Paris PSL) 17/10/2024, Elouan Argouarc'h (TelecomSudParis/CEA) 11/12/2024

11.3.1 Productions (articles, videos, podcasts, serious games, ...)

• Alessandro Leite and Marc Schoenauer produced the Webinar Memetic Semantic Genetic Programming, within the TRUST-AI Horizon Europe project
• Michèle Sebag participated in Ce qui échappe à l'intelligence artificielle, eds. François Levin, Étienne Ollion. ISBN : 9791037038449
• Michèle Sebag: conf.de presse IA et mésinformation (with Nicolas Curien); report on IA et mésinformation, ed. N. Curien, Académie des Technologies, 13/12/24; report on Prouesses et limites de l'imittion artificielle de langages, ed G. Roucayrol, https://www.academie-technologies.fr/publications/prouesses-et-limites-de-limitation-artificielle-de-langages-avis/

International journals

• 9 articleY.Yvenn Amara-Ouali, Y.Yannig Goude, N.Nathan Doumèche, P.Pascal Veyret, A.Alexis Thomas, D.Daniel Hebenstreit, T.Thomas Wedenig, A.Arthur Satouf, A.Aymeric Jan, Y.Yannick Deleuze, P.Paul Berhaut, S.Sébastien Treguer and T.Tiphaine Phe-Neau. Forecasting Electric Vehicle Charging Station Occupancy: Smarter Mobility Data Challenge.Journal of Data-centric Machine Learning Research1July 2024HALback to text
• 10 articleI.Igor Carrara, B.Bruno Aristimunha, M.-C.Marie-Constance Corsi, R. Y.Raphael Yokoingawa de Camargo, S.Sylvain Chevallier and T.Théodore Papadopoulo. Geometric Neural Network based on Phase Space for BCI decoding.Journal of Neural EngineeringOctober 2024HALDOIback to text
• 11 articleG.Giovanni Catania, A.Aurélien Decelle and B.Beatriz Seoane. The Copycat Perceptron: Smashing Barriers Through Collective Learning.Physical Review E 10962024, 065313HALDOIback to text
• 12 articleR.Rahul Chacko, F.François Landes, G.Giulio Biroli, O.Olivier Dauchot, A.Andrea Liu and D.David Reichman. Dynamical Facilitation Governs the Equilibration Dynamics of Glasses.Physical Review X143July 2024, 031012HALDOIback to text
• 13 articleM.-C.Marie-Constance Corsi, P.Pierpaolo Sorrentino, D. P.Denis P Schwartz, N.Nathalie George, L. L.Leonardo L. Gollo, S.Sylvain Chevallier, L.Laurent Hugueville, A. E.Ari E Kahn, S.Sophie Dupont, D. S.Danielle S Bassett, V.Viktor Jirsa and F.Fabrizio de Vico Fallani. Measuring Neuronal Avalanches to inform Brain-Computer Interfaces.iScience2712024, 108734HALDOIback to text
• 14 articleA.Aurélien Decelle, C.Cyril Furtlehner, A. d.Alfonso de Jesús Navas Gómez and B.Beatriz Seoane. Inferring effective couplings with Restricted Boltzmann Machines.SciPost Physics164April 2024, 095HALDOIback to text
• 15 articleD.Diego Delle Donne, M.Matthieu Kowalski and L.Leo Liberti. A Novel Integer Linear Programming Approach for Global L0 Minimization.Journal of Machine Learning Research241March 2024, Art. no. 382, p.18322-18349HALDOIback to text
• 16 articleG.Gerhard Jung, R. M.Rinske M Alkemade, V.Victor Bapst, D.Daniele Coslovich, L.Laura Filion, F. P.François P. Landes, A.Andrea Liu, F. S.Francesco Saverio Pezzicoli, H.Hayato Shiba, G.Giovanni Volpe, F.Francesco Zamponi, L.Ludovic Berthier and G.Giulio Biroli. Roadmap on machine learning glassy liquids.Nature Rev.Phys.722025, 91-104HALDOIback to textback to text
• 17 articleB.Bruna Junqueira, B.Bruno Aristimunha, S.Sylvain Chevallier and R.Raphael de Camargo. A systematic evaluation of Euclidean alignment with deep learning for EEG decoding.Journal of Neural Engineering213May 2024, 036038HALDOIback to text
• 18 articleM.Maël Lefeuvre, M. D.Michael David Martin, F.Flora Jay, M.-C.Marie-Claude Marsolier and C.Céline Bon. GRUPS-rs, a high-performance ancient DNA genetic relatedness estimation software relying on pedigree simulations.Human Population Genetics and Genomics41January 2024, 0001HALDOIback to text
• 19 articleA.Alessandro Leite and M.Marc Schoenauer. Memetic Semantic Boosting for Symbolic Regression.Genetic Programming and Evolvable Machines2025. In press. HALback to text
• 20 articleC.Céline Loot, G. A.Gael A Millot, E.Egill Richard, E.Eloi Littner, C.Claire Vit, F.Frédéric Lemoine, B.Bertrand Néron, J.Jean Cury, B.Baptiste Darracq, T.Théophile Niault, D.Delphine Lapaillerie, V.Vincent Parissi, E. P.Eduardo P C Rocha and D.Didier Mazel. Integron cassettes commonly integrate into bacterial genomes via widespread non-classical attG sites.Nature Microbiology91January 2024, 228-240HALDOI
• 21 articleE.Emmanuel Menier, S.Sebastian Kaltenbach, M.Mouadh Yagoubi, M.Marc Schoenauer and P.Petros Koumoutsakos. Interpretable learning of effective dynamics for multiscale systems.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences4812306January 2025, 20240167HALDOIback to textback to text
• 22 articleM.Matthieu Nastorg, M. A.Michele Alessandro Bucci, T.Thibault Faney, J.-M.Jean-Marc Gratien, G.Guillaume Charpiat and M.Marc Schoenauer. An implicit GNN solver for Poisson-like problems.Computers & Mathematics with Applications176December 2024, 270-288HALDOIback to text
• 23 articleM. H.Manh Hung Nguyen, L.Lisheng Sun-Hosoya and I.Isabelle Guyon. Meta-Learning from Learning Curves for Budget-Limited Algorithm Selection.Pattern Recognition Letters185September 2024, 225-231HALDOIback to text
• 24 articleF. S.Francesco Saverio Pezzicoli, G.Guillaume Charpiat and F. P.François Pascal Landes. Rotation-equivariant graph neural networks for learning glassy liquids representations.SciPost Physics165May 2024, 136HALDOIback to text
• 25 articleV. M.Vincenzo Maria Schimmenti, G.Giuseppe Petrillo, A.Alberto Rosso and F. P.François P. Landes. Assessing the predicting power of GPS data for aftershocks forecasting.Seismological Research Letters956November 2024, 3243–3249HALDOIback to text
• 26 articleM.Manon Verbockhaven, T.Théo Rudkiewicz, S.Sylvain Chevallier and G.Guillaume Charpiat. Growing Tiny Networks: Spotting Expressivity Bottlenecks and Fixing Them Optimally.Transactions on Machine Learning Research JournalOctober 2024HALback to textback to text

International peer-reviewed conferences

• 27 inproceedingsN.Nicolas Atienza, R.Roman Bresson, C.Cyriaque Rousselot, P.Philippe Caillou, J.Johanne Cohen, C.Christophe Labreuche and M.Michèle Sebag. Cutting the Black Box: Conceptual Interpretation of a Deep Neural Net with Multi-Modal Embeddings and Multi-Criteria Decision Aid.Proceedings of the Thirty-Third International Joint Conference on Artificial IntelligenceIJCAI-24, Thirty-Third International Joint Conference on Artificial IntelligenceJeju, South KoreaInternational Joint Conferences on Artificial Intelligence Organization2024, 3669-3678HALDOIback to textback to textback to text
• 28 inproceedingsN.Nicolas Atienza, C.Christophe Labreuche, J.Johanne Cohen and M.Michèle Sebag. Provably Safeguarding a Classifier from OOD and Adversarial Samples: an Extreme Value Theory Approach.Proc. ICLR'25ICLR 2025 - The Thirteenth International Conference on Learning RepresentationsSingapore (SG), SingaporeJanuary 2025HALback to text
• 29 inproceedingsD.Dimitrios Bachtis, G.Giulio Biroli, A.Aurélien Decelle and B.Beatriz Seoane. Cascade of phase transitions in the training of Energy-based models.NeurIPS2024Vancouver, United StatesarXiv2024HALDOIback to textback to text
• 30 inproceedingsN.Nicolas Bereux, A.Aurélien Decelle, C.Cyril Furtlehner, L.Lorenzo Rosset and B.Beatriz Seoane. Fast training and sampling of Restricted Boltzmann Machines.13th International Conference on Learning Representations - ICLR 2025Singaour, SingaporeMarch 2025HALback to textback to text
• 31 inproceedingsE.Eva Boguslawski, A.Alessandro Leite, B.Benjamin Donnot, M.Matthieu Dussartre and M.Marc Schoenauer. Emulation of Zonal Controllers for the Power System Transport Problem.ML4SPS 2024 - Machine Learning for Sustainable Power Systems ECML 2024 WorkshopVilnius, LithuaniaSeptember 2024HALback to text
• 32 inproceedingsS.Shuyu Dong, M.Michèle Sebag, K.Kento Uemura, A.Akito Fujii, S.Shuang Chang, Y.Yusuke Koyanagi and K.Koji Maruhashi. DCDILP: a distributed learning method for large-scale causal structure learning.Proc. AAAI 2025AAAI 25 - The 39th Annual AAAI Conference on Artificial IntelligencePhiladelphia (PA), United StatesFebruary 2025HALback to textback to text
• 33 inproceedingsS.Stella Douka, M.Manon Verbockhaven, T.Théo Rudkiewicz, S.Stéphane Rivaud, F. P.François P. Landes, S.Sylvain Chevallier and G.Guillaume Charpiat. Growth strategies for arbitrary DAG neural architectures.Proc. ESANN'25ESANN 2025 - 33th European Symposium on Artificial Neural NetworksBruges, BelgiumApril 2025HALback to text
• 34 inproceedingsA.Alice Lacan, B.Blaise Hanczar and M.Michele Sebag. Frugal Generative Modeling for Tabular Data.Lecture Notes in Computer ScienceECML PKDD 2024 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases14948Lecture Notes in Computer ScienceVilnius, LithuaniaSpringer Nature SwitzerlandAugust 2024, 55--72HALDOIback to text
• 35 inproceedingsJ.-B.Jean-Baptiste Malagnoux and M.Matthieu Kowalski. From Convolutional Sparse Coding To *-NMF Factorization of Time-Frequency Coefficients.2024 IEEE International Conference on Acoustics, Speech and Signal ProcessingICASSP 2024 - IEEE International Conference on Acoustics, Speech and Signal ProcessingICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)Séoul, South KoreaIEEE2024, 5530-5534HALDOIback to text
• 36 inproceedingsA.Apolline Mellot, A.Antoine Collas, S.Sylvain Chevallier, D.Denis Engemann and A.Alexandre Gramfort. Physics-informed and Unsupervised Riemannian Domain Adaptation for Machine Learning on Heterogeneous EEG Datasets.EUSIPCO 2024 - 32nd European Signal Processing ConferenceLyon, FrancearXiv2024HALDOIback to text
• 37 inproceedingsA.Apolline Mellot, A.Antoine Collas, S.Sylvain Chevallier, A.Alexandre Gramfort and D.Denis Engemann. Geodesic optimization for predictive shift adaptation on EEG data.Proc. NeuIPS'24NeurIPS 2024 - 38th Conference on Neural Information Processing SystemsVancouver, CanadaDecember 2024HALback to textback to text
• 38 inproceedingsT.Thibault Monsel, O.Onofrio Semeraro, L.Lionel Mathelin and G.Guillaume Charpiat. Time and State Dependent Neural Delay Differential Equations.PMLRML-DE@ECAI 2024 : Machine Learning Meets Differential Equations: From Theory to ApplicationsSantiago de compostela, Galicia, SpainOctober 2024HALback to text
• 39 inproceedingsM.Matthieu Nastorg, J.-M.Jean-Marc Gratien, T.Thibault Faney, M. A.Michele Alessandro Bucci, G.Guillaume Charpiat and M.Marc Schoenauer. Multi-Level GNN Preconditioner for Solving Large Scale Problems.2024 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)San Francisco (CA, USA), United StatesIEEE Computer SocietyMay 2024, 486-495HALDOIback to text
• 40 inproceedingsA.Audrey Poinsot, A.Alessandro Leite, N.Nicolas Chesneau, M.Michele Sebag and M.Marc Schoenauer. Learning Structural Causal Models through Deep Generative Models: Methods, Guarantees, and Challenges.Proc. Thirty-Third International Joint Conference on Artificial IntelligenceIJCAI 2024 - Survey TrackProceedings of the Thirty-Third International Joint Conference on Artificial IntelligenceJeju, South KoreaAugust 2024, 8207-8215HALDOIback to textback to text
• 41 inproceedingsG. H.Gustavo H Rodrigues, B.Bruno Aristimunha, S.Sylvain Chevallier and R. Y.Raphael Y de Camargo. Combining Euclidean Alignment and Data Augmentation for BCI decoding.EUSIPCO 2024 - 32nd European Signal Processing ConferenceLyon, France2024HALDOIback to text
• 42 inproceedingsN.Nilo Schwencke and C.Cyril Furtlehner. ANAGRAM: a natural gradient relative to adapted model for efficient PINNS learning.In proceeding of ICLR 2025ICLR 2025 - International Conference on Learning Representations13th International Conference on Learning Representations (ICLR 2025)Singapour, SingaporeApril 2025HALback to textback to text
• 43 inproceedingsD.Dylan Sechet, F.Francesca Bugiotti, M.Matthieu Kowalski, E.Edouard D’hérouville and F.Filip Langiewicz. A Hierarchical Deep Learning Approach for Minority Instrument Detection.DAFx 2024 - the 27th International Conference on Digital Audio EffectsGuildford Surrey, United KingdomSeptember 2024HALback to text
• 44 inproceedingsT.Thibault de Surrel, S.Sylvain Chevallier, F.Fabien Lotte and F.Florian Yger. Averaging trajectories on the manifold of symmetric positive definite matrices.EUSIPCO 2024 - 32nd European Signal Processing ConferenceLyon, FranceAugust 2024HALback to text
• 45 inproceedingsA.Antoine Szatkownik, C.Cyril Furtlehner, G.Guillaume Charpiat, B.Burak Yelmen and F.Flora Jay. Towards creating longer genetic sequences with GANs: Generation in principal component space.PMLRMLCB 2023 - 18th Conference on Machine Learning in Computational BiologySeattle, United States2024HALback to text
• 46 inproceedingsS.Sébastien Velut, S.Sylvain Chevallier, M.-C.Marie-Constance Corsi and F.Frédéric Dehais. Deep Riemannian Neural Architectures for Domain Adaptation in Burst cVEP-based Brain Computer Interface.ESANN 2024 - 32nd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine LearningBruges, BelgiumCiaco - i6doc.comOctober 2024, 571-576HALDOIback to text
• 47 inproceedingsJ.Jonathan Xu, B.Bruno Aristimunha, M. E.Max Emanuel Feucht, E.Emma Qian, C.Charles Liu, T.Tazik Shahjahan, M.Martyna Spyra, S. Z.Steven Zifan Zhang, N.Nicholas Short, J.Jioh Kim, P.Paula Perdomo, R. R.Ricky Renfeng Mao, Y.Yashvir Sabharwal, M.Michael Ahedor, M.Moaz Shoura and A.Adrian Nestor. Alljoined1 -A dataset for EEG-to-Image decoding.Workshop Data Curation and Augmentation in Medical Imaging at 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)Settle, United StatesarXivMay 2024HALDOIback to text
• 48 inproceedingsM.Mouadh Yagoubi, D.David Danan, M.Milad Leyli-Abadi, J.-P.Jean-Patrick Brunet, M.Maroua Gmati, A.Ahmed Mazari, F.Florent Bonnet, A.Asma Farjallah, P.Paola Cinnella, P.Patrick Gallinari, M.Marc Schoenauer and A.Asma Farjallah. NeurIPS 2024 ML4CFD Competition: Harnessing Machine Learning for Computational Fluid Dynamics in Airfoil Design.NeurIPS 2024 - The 38th Annual Conference on Neural Information Processing SystemsVancouver, CanadaDecember 2024HALback to textback to text

Conferences without proceedings

• 49 inproceedingsP. H.Paulo Henrique Couto, Q. P.Quang Phuoc Ho, N.Nageeta Kumari, B. K.Benedictus Kent Rachmat, T. G.Thanh Gia Hieu Khuong, I.Ihsan Ullah and L.Lisheng Sun-Hosoya. RelevAI-Reviewer: A Benchmark on AI Reviewers for Survey Paper Relevance.CAp 2024 - Conférence sur l'Apprentissage AutomatiqueLille, FranceJuly 2024HALback to text

Doctoral dissertations and habilitation theses

• 50 thesisG.Guillaume Bied. Designing Recommender Systems for the Labor Market.Université Paris-SaclayJuly 2024HALback to text
• 51 thesisR.Romain Egele. Optimization of Learning Workflows at Large Scale on High-Performance Computing Systems.Université Paris-SaclayJune 2024HALback to text
• 52 thesisA.Armand Lacombe. Changes of representation for counter-factual inference.Université Paris-SaclayMarch 2024HALback to text
• 53 thesisA.Apolline Mellot. Machine learning and domain adaptation for enhancing the measure of brain health with MEG and EEG signals.Université Paris-SaclayNovember 2024HAL
• 54 thesisE.Emmanuel Menier. Deep Learning for Reduced Order Modeling.Université Paris-SaclayJanuary 2024HALback to text
• 55 thesisM.Matthieu Nastorg. Scalable GNN Solutions for CFD Simulations.Université Paris-SaclayApril 2024HALback to text
• 56 thesisF.Francesco Pezzicoli. Statistical Physics - Machine Learning Interplay : from Addressing Class Imbalance with Replica Theory to Predicting Dynamical Heterogeneities with SE(3)-equivariant Graph Neural Networks.Université Paris-SaclayDecember 2024HALback to textback to text

Reports & preprints

• 57 miscP.Pierre Barbault, M.Matthieu Kowalski and C.Charles Soussen. Bernoulli and Gauss Take a Look at the MAP.November 2024HALback to text
• 58 miscP.Pierre Barbault, M.Matthieu Kowalski and C.Charles Soussen. LEMUR: Latent EM Unsupervised Regression for Sparse Inverse Problems.April 2024HALback to text
• 59 miscS.Sylvain Chevallier, I.Igor Carrara, B.Bruno Aristimunha, P.Pierre Guetschel, S.Sara Sedlar, B.Bruna Junqueira Lopes, S.Sébastien Velut, S.Salim Khazem and T.Thomas Moreau. The largest EEG-based BCI reproducibility study for open science: the MOABB benchmark.April 2024HAL
• 60 miscR.Romain Egele, F.Felix Mohr, T.Tom Viering and P.Prasanna Balaprakash. The Unreasonable Effectiveness Of Early Discarding After One Epoch In Neural Network Hyperparameter Optimization.April 2024HALback to text
• 61 miscD.Dilek Koptekin, A.Ayça Aydoğan, C.Cansu Karamurat, N. E.N. Ezgi Altınışık, K. B.Kıvılcım Başak Vural, D. D.D. Deniz Kazancı, A. K.Ayça Küçükakdağ Doğu, D.Damla Kaptan, H. C.Hasan Can Gemici, E.Eren Yüncü, G.Gülsün Umurtak, R.Refik Duru, E.Erkan Fidan, Ö.Özlem Çevik, B.Burçin Erdoğu, T.Taner Korkut, C.Christopher Knüsel, S.Scott Haddow, C. S.Clark Spencer Larsen, R.Rana Özbal, F.Fokke Gerritsen, E.Eylem Özdoğan, U. O.Uygar Ozan Usanmaz, Y. C.Yasin Cemre Derici, M.Mine Uçmazoğlu, F.Flora Jay, M.Mehmet Özdoğan, A.Anders Götherström, Y. S.Yılmaz Selim Erdal, A.-S.Anna-Sapfo Malaspinas, Ç.Çiğdem Atakuman, F.Füsun Özer and M.Mehmet Somel. Out-of-Anatolia: cultural and genetic interactions during the Neolithic expansion in the Aegean.June 2024HALDOIback to text
• 62 miscM.Matthieu Kowalski, B.Benoît Malézieux, T.Thomas Moreau and A.Audrey Repetti. Analysis and synthesis denoisers for forward-backward plug-and-play algorithms.December 2024HALback to text
• 63 miscB.Benoît Malézieux, F.Florent Michel, T.Thomas Moreau and M.Matthieu Kowalski. Where prior learning can and can't work in unsupervised inverse problems.November 2024HALback to text
• 64 miscT.Thibault Monsel, E.Emmanuel Menier, L.Lionel Mathelin, O.Onofrio Semeraro and G.Guillaume Charpiat. Neural DDEs with Learnable Delays for Partially Observed Dynamical Systems.October 2024HALback to text
• 65 miscA.Adrien Pavão. Hands-on tutorial on how to create your own challenge or benchmark.January 2024HALback to text
• 66 miscM.Michele Quattromini, M. A.Michele Alessandro Bucci, S.Stefania Cherubini and O.Onofrio Semeraro. Graph Neural Networks and Differential Equations: A hybrid approach for data assimilation of fluid flows.November 2024HALback to text
• 67 miscA.Antoine Szatkownik, C.Cyril Furtlehner, G.Guillaume Charpiat, B.Burak Yelmen and F.Flora Jay. Latent generative modeling of long genetic sequences with GANs.August 2024HALDOI
• 68 miscA.Antoine Szatkownik, L.Léo Planche, M.Maïwen Demeulle, T.Titouan Chambe, M.María Ávila-Arcos, E.Emilia Huerta-Sanchez, C.Cyril Furtlehner, C.Cyril Furtlehner, F.Flora Jay and B.Burak Yelmen. Diffusion-based artificial genomes and their usefulness for local ancestry inference.October 2024HALDOIback to text
• 69 reportS.Sebastien Treguer and M.Marc Schoenauer. Lessons learned from TAILOR benchmarks/challenges.INRIA - team TAUDecember 2024HALback to text
• 70 miscW.-W.Wei-Wei Tu and A.Adrien Pavão. Special designs and competition protocols.January 2024HALback to text
• 71 miscM.Mathurin Videau, M.Mariia Zameshina, A.Alessandro Leite, L.Laurent Najman, M.Marc Schoenauer and O.Olivier Teytaud. Evolutionary Retrofitting.October 2024HALback to text
• 72 miscB.Burak Yelmen, M.Maris Alver, E. B.Estonian Biobank Research Team, F.Flora Jay and L.Lili Milani. Interpreting artificial neural networks to detect genome-wide association signals for complex traits.2024HALDOIback to text
• 73 miscM.Mariia Zameshina, M.Mathurin Videau, A.Alessandro Leite, M.Marc Schoenauer, L.Laurent Najman and O.Olivier Teytaud. Agnostic latent diversity enhancement in generative modeling.October 2024HALback to text

Other scientific publications

• 74 inproceedingsB.Bruno Aristimunha, T.Thomas Moreau, S.Sylvain Chevallier, R. Y.Raphael Y de Camargo and M.-C.Marie-Constance Corsi. What is the best model for decoding neurophysiological signals? Depends on how you evaluate.CNS 2024 - 33rd Annual Computational Neuroscience MeetingNatal, BrazilJuly 2024HAL