2025​​Activity reportProject-TeamMALICE​​​‌

7.1.1‌​‌ SwiftHohenbergPseudoSpectral

• Keywords:
  Partial differential​​ equation, Numerical solver, Self-organization​​​‌
• Functional Description:

  This software‌ solves the dimensionless Swift-Hohenberg‌​‌ Equation with a cubic-quadratic​​ nonlinear term. The Swift-Hohenberg​​​‌ Equation is a widely‌ studied model for pattern‌​‌ formation, derived using symmetry​​ arguments. As such, it​​​‌ serves as a maximally‌ symmetric model for systems‌​‌ exhibiting spontaneous pattern formation,​​ making it broadly applicable​​​‌ to various physical contexts.‌

  The solver employs pseudospectral‌​‌ methods based on Fourier​​ representations to ensure efficient​​​‌ and accurate computation. Importantly,‌ the solver is physically‌​‌ constrained, using a Lyapunov​​ functional to validate solutions,​​​‌ ensuring that the computed‌ results remain consistent with‌​‌ the physical behavior expected​​ from the system.

  Designed​​​‌ with versatility and automation‌ in mind, the solver‌​‌ is capable of generating​​ a large number of​​​‌ solutions with minimal supervision.‌ This feature makes it‌​‌ particularly well-suited for creating​​ synthetic datasets to be​​​‌ used in Physics Guided‌ Machine Learning. Specifically, it‌​‌ addresses problems related to​​ the femtosecond laser-induced self-organization​​​‌ of matter, where data‌ augmentation is crucial for‌​‌ improving the performance and​​ generalization of machine learning​​​‌ models.

• Publications:
  ujm-04157829,‌ ujm-03823722, tel-04515418v1
• Contact:‌​‌
  Eduardo Brandao

7.1.2 GUAP​​

• Name:
  Generalized Universal Adversarial​​​‌ Perturbations
• Keyword:
  Machine learning‌
• Functional Description:
  The software‌​‌ extends universal attacks by​​ jointly learning a set​​​‌ of perturbations to choose‌ from, aiming to maximize‌​‌ the success rate of​​ attacks against deep neural​​​‌ network models.
• URL:
  https://­github.­com/­JordanFrecon/­guap‌
• Contact:
  Jordan Frecon Patracone‌​‌
• Partner:
  LITIS

7.1.3 GraSPy​​

• Name:
  Graph Signal Processing​​​‌ for Python
• Keywords:
  Graph,‌ Signal processing, Machine learning‌​‌
• Functional Description:
  Implements classic​​ components of such a​​​‌ toolbox, including (weighted) graph‌ generalization and transformation and‌​‌ of their signals (a.k.a.​​ vertex attributes), specific data​​​‌ visualization using Plotly, signal‌ transformation with spectral methods‌​‌ (using IAGFTs), and graph​​ learning.
• URL:
  https://­gitlab.­inria.­fr/­begiraul/­joint-graph-learning
• Publication:​​​‌
  hal-04025968
• Contact:
  Benjamin Girault‌

7.1.4 Scikit-SpLearn

• Name:
  Toolbox‌​‌ for the spectral learning​​ of weighted automata
• Keywords:​​​‌
  Machine learning, Weighted automata‌
• Scientific Description:
  The core‌​‌ idea of the spectral​​ learning of weighted automata​​​‌ is to use a‌ rank factorization of a‌​‌ complete sub-block of the​​ Hankel matrix of a​​​‌ target series to induce‌ a weighted automaton.
• Functional‌​‌ Description:
  The toolkit contains​​​‌ an efficient implementation of​ weighted automata, a library​‌ to transformed any dataset​​ containing strings of different​​​‌ length into the scitkit-learn​ format for data, 4​‌ variants of the spectral​​ learning algorithm.
• URL:
  https://­remieyraud.­github.­io/­scikit-splearn/​​​‌
• Publication:
  hal-01777169
• Contact:
  Rémi​ Eyraud
• Partner:
  Laboratoire d'Informatique​‌ et des Systèmes (LIS)​​ Université Aix-Marseille

7.1.5 SoundScapeExplorer​​​‌

• Keywords:
  Acoustics, Data analysis,​ Data visualization, Soundscape
• Functional​‌ Description:

  SSE stands for​​ SoundScapeExplorer and works with​​​‌ passive acoustic monitoring campaigns​ made of several microphones​‌ recording during hours, days​​ or week.

  It analyzes​​​‌ these data by extracting​ various features and indicators​‌ for each signal window​​ (typically 1 second) then​​​‌ aggregating them by chunks​ (e.g. from 5 seconds​‌ to 1 or more​​ minutes). Using a selection​​​‌ of dimensionality reduction techniques,​ it allows to visualize​‌ the whole campaign as​​ a point cloud where​​​‌ each point corresponds to​ a temporal chunk on​‌ one microphone. The visualization​​ is interactive, allowing coloring​​​‌ by criteria, filtering, listening​ to audio data, cluster​‌ analysis and beyond.

• URL:​​
  https://­sound-scape-explorer.­github.­io/
• Contact:
  Rémi Emonet​​​‌
• Partner:
  Laboratoire ENES

8.1.1 On​‌ the Closed-Form of Flow​​ Matching: Generalization Does Not​​​‌ Arise from Target Stochasticity​ (NeurIPS'25) 2

Participants: Quentin​‌ Bertrand, Rémi Emonet​​.

Collaboration with Anne​​​‌ Gagneux and Mathurin Massias​ (Inria OCKHAM)

Modern deep​‌ generative models can now​​ produce high-quality synthetic samples​​​‌ that are often indistinguishable​ from real training data.​‌ A growing body of​​ research aims to understand​​​‌ why recent methods–such as​ diffusion and flow matching​‌ techniques–generalize so effectively. Among​​ the proposed explanations are​​​‌ the inductive biases of​ deep learning architectures and​‌ the stochastic nature of​​ the conditional flow matching​​​‌ loss. In this work,​ we rule out the​‌ latter -the noisy nature​​ of the loss -as​​​‌ a primary contributor to​ generalization in flow matching.​‌ First, we empirically show​​ that in high-dimensional settings,​​​‌ the stochastic and closed-form​ versions of the flow​‌ matching loss yield nearly​​ equivalent losses. Then, using​​​‌ state-of-the-art flow matching models​ on standard image datasets,​‌ we demonstrate that both​​ variants achieve comparable statistical​​​‌ performance, with the surprising​ observation that using the​‌ closed-form can even improve​​ performance.

Keywords: Generative AI,​​ Flow Matching

8.1.2 A​​​‌ Visual Dive into Conditional‌ Flow Matching (ICLR Blogposts'25)‌​‌ 4

Participants: Quentin Bertrand​​, Rémi Emonet.​​​‌

Collaboration with Ségolène Martin‌ (Technische Universität Berlin), Anne‌​‌ Gagneux and Mathurin Massias​​ (Inria OCKHAM)

Conditional flow​​​‌ matching (CFM) was introduced‌ by three simultaneous papers‌​‌ at ICLR 2023, through​​ different approaches (conditional matching,​​​‌ rectifying flows and stochastic‌ interpolants). The main part‌​‌ of this post, Section​​ 2, explains CFM by​​​‌ using both visual intuitions‌ and insights on its‌​‌ probabilistic formulations. Section 1​​ introduces nomalizing flows; it​​​‌ can be skipped by‌ reader familiar with the‌​‌ topic, or that wants​​ to cover them later.​​​‌ Section 3 opens on‌ the links between CFM‌​‌ and other approaches, and​​ ends with a 'CFM​​​‌ playground'.

Keywords: Generative AI,‌ Flow Matching

8.2.1 A‌ Bregman Proximal Viewpoint on‌​‌ Neural Operators (ICML'25) 6​​

Participants: Abdel-Rahim Mezidi,​​​‌ Jordan Patracone, Amaury‌ Habrard, Rémi Emonet‌​‌, Marc Sebban.​​

Collaboration with Saverio Salzo​​​‌ (Sapienza University of Rome,‌ Italy)

We present several‌​‌ advances on neural operators​​ by viewing the action​​​‌ of operator layers as‌ the minimizers of Bregman‌​‌ regularized optimization problems over​​ Banach function spaces. The​​​‌ proposed framework allows interpreting‌ the activation operators as‌​‌ Bregman proximity operators from​​ dual to primal space.​​​‌ This novel viewpoint is‌ general enough to recover‌​‌ classical neural operators as​​ well as a new​​​‌ variant, coined Bregman neural‌ operators, which includes the‌​‌ inverse activation operator and​​ features the same expressivity​​​‌ of standard neural operators.‌ Numerical experiments support the‌​‌ added benefits of the​​ Bregman variant of Fourier​​​‌ neural operators for training‌ deeper and more accurate‌​‌ models.

Keywords: Neural Operators,​​ Physics-informed ML

8.2.2 Provably​​​‌ Accurate Adaptive Sampling for‌ Collocation Points in Physics-informed‌​‌ Neural Networks (ECML'25) 3​​

Participants: Antoine Caradot,​​​‌ Rémi Emonet, Abdel-Rahim‌ Mezidi, Amaury Habrard‌​‌, Marc Sebban.​​

Despite considerable scientific advances​​​‌ in numerical simulation, efficiently‌ solving PDEs remains a‌​‌ complex and often expensive​​ problem. Physics-informed Neural Networks​​​‌ (PINN) have emerged as‌ an efficient way to‌​‌ learn surrogate solvers by​​ embedding the PDE in​​​‌ the loss function and‌ minimizing its residuals using‌​‌ automatic differentiation at so-called​​ collocation points. Originally uniformly​​​‌ sampled, the choice of‌ the latter has been‌​‌ the subject of recent​​ advances leading to adaptive​​​‌ sampling refinements for PINNs.‌ In this work, leveraging‌​‌ a new quadrature method​​ for approximating definite integrals,​​​‌ we introduce a provably‌ accurate sampling method for‌​‌ collocation points based on​​ the Hessian of the​​​‌ PDE residuals. Comparative experiments‌ conducted on a set‌​‌ of 1D and 2D​​ PDEs demonstrate the benefits​​​‌ of our method.

Keywords:‌ PINNS, Collocation points, Physics-informed‌​‌ ML

8.2.3 Conformal Online​​​‌ Learning of Deep Koopman​ Linear Embeddings (NeurIPS'25) 5​‌

Participants: Ben Gao,​​ Jordan Patracone.

Collaboration​​​‌ with Stéphane Chrétien (ERIC-Lyon​ 2) and Olivier Alata​‌ (LabHC)

We introduce in​​ this work Conformal Online​​​‌ Learning of Koopman embeddings​ (COLoKe), a novel framework​‌ for adaptively updating Koopman-invariant​​ representations of nonlinear dynamical​​​‌ systems from streaming data.​ Our modeling approach combines​‌ deep feature learning with​​ multi-step prediction consistency in​​​‌ the lifted space, where​ the dynamics evolve linearly.​‌ To prevent overfitting, COLoKe​​ employs a conformal-style mechanism​​​‌ that shifts the focus​ from evaluating the conformity​‌ of new states to​​ assessing the consistency of​​​‌ the current Koopman model.​ Updates are triggered only​‌ when the current model's​​ prediction error exceeds a​​​‌ calibrated threshold, allowing selective​ refinement of the Koopman​‌ operator and embedding. Empirical​​ results on benchmark dynamical​​​‌ systems demonstrate the effectiveness​ of COLoKe in maintaining​‌ long-term predictive accuracy while​​ significantly reducing unnecessary updates.​​​‌

Keywords: Nonlinear dynamical systems,​ Streaming data.

8.2.4 Using​‌ Signatures and Koopman Operator​​ to Learn Nonlinear Dynamics​​​‌ (GSI'25)

Participants: Ben Gao​, Jordan Patracone.​‌

Collaboration with Stéphane Chrétien​​ (ERIC-Lyon 2) and Olivier​​​‌ Alata (LabHC)

We propose​ a novel framework for​‌ predicting the evolution of​​ dynamical systems by learning​​​‌ the Koopman operator in​ the space of linear​‌ functionals on the Signature​​ transform of trajectory data.​​​‌ The Signature, a central​ object in rough path​‌ theory, provides a universal​​ and compact representation of​​​‌ paths through iterated integrals,​ enabling linear models to​‌ approximate a wide class​​ of non-linear functionals. By​​​‌ restricting observables to lie​ in the span of​‌ truncated Signatures, we construct​​ a finite-dimensional approximation of​​​‌ the Koopman operator, which​ we estimate directly from​‌ data using regularized linear​​ regression. This approach merges​​​‌ the expressiveness of operator-theoretic​ methods with the structural​‌ richness of Signature features.​​

Keywords: Nonlinear dynamical systems,​​​‌ Signature transform.

8.2.5 Contextual​ Hypernetwork for Adaptive Prediction​‌ of Laser-Induced Colors on​​ Quasi-Random Plasmonic Metasurfaces (ECML'25)​​​‌ 25

Participants: Thibault Girardin​, Amaury Habrard.​‌

Collaboration with Nathalie Destouches​​ (LabHC)

Laser processing is​​​‌ a rapid, versatile, and​ low-cost technology to print​‌ images on large surfaces.​​ When applied to very​​​‌ thin films embedded with​ disordered metallic nanoparticles, known​‌ as quasi-random plasmonic metasurfaces,​​ it generates colors that​​​‌ vary with the observation​ mode, making it valuable​‌ for visual security applications.​​ Predicting these colors in​​​‌ different modes from the​ knowledge of laser processing​‌ parameters and the initial​​ state of the metasurface​​​‌ can accelerate the industrialization​ process. However, there is​‌ no general physical model​​ able to make this​​​‌ prediction accurately in various​ modes. In order to​‌ address this issue, this​​ paper proposes a data-driven​​​‌ approach for learning deep​ models on experimental data​‌ able to predict the​​ colors observed in different​​​‌ environments for a large​ range of laser processing​‌ parameters. We leverage a​​ framework that learns jointly​​​‌ a shared latent space​ for multiple environments together​‌ with a contextual representation​​ specific to each. This​​​‌ contextual representation is generated​ by an hypernetwork conditioned​‌ on an interpretable context​​ vector. This context vector​​ can be learned from​​​‌ few data allowing fast‌ adaptation to new environments.‌​‌ This approach demonstrates that​​ a single model can​​​‌ learn to predict a‌ large range of colors‌​‌ across different environments. Its​​ effectiveness is demonstrated through​​​‌ its ability to rapidly‌ adapt to new scenarios‌​‌ with minimal data and​​ to serve as an​​​‌ improved weight initializer for‌ finetuning when larger datasets‌​‌ are available. Source code​​ and datasets are available​​​‌ on Gitlab.

Keywords: Laser-matter‌ interaction, Domain Adaptation, Hypernetwork‌​‌

8.2.6 Physics-Informed Machine Learning​​ for Modeling CO2 Capture​​​‌ from Scarce Data (ICTAI'25)‌ 18

Participants: Mickael Gault‌​‌, Rémi Emonet,​​ Marc Sebban.

Collaboration​​​‌ with P. Bachaud, B.‌ Celse (IFPEN, France)

Accurate‌​‌ modeling of complex industrial​​ processes often relies on​​​‌ costly mechanistic simulations grounded‌ in physical principles. In‌​‌ this paper, we investigate​​ the subject of CO2​​​‌ capture, a major environmental‌ challenge, through the absorption‌​‌ column of an amine-based​​ post-combustion process. The modeling​​​‌ of such unit at‌ industrial scale faces two‌​‌ difficulties: (i) theoretical models,​​ efficient at laboratory scale,​​​‌ might fail to fully‌ reflect the complexity of‌​‌ the numerous intertwined phenomena​​ occurring in the absorber,​​​‌ (ii) the cost and‌ uncertainty of industrial observation‌​‌ data make purely data-driven​​ approaches unfeasible.

To tackle​​​‌ both this low data‌ regime and inaccurate physical‌​‌ models, we envision this​​ CO2 capture problem through​​​‌ the lens of Physics-informed‌ Machine Learning (PiML). We‌​‌ present a hybrid (data+knowledge)​​ model where the scarce​​​‌ observation data complement the‌ physical model, while the‌​‌ latter ensures that the​​ predictions remain physically consistent.​​​‌ Beyond the standard use‌ of simulation data for‌​‌ learning and the embedding​​ of physical laws as​​​‌ regularization, the originality of‌ our PiML algorithm compared‌​‌ to other methods in​​ the literature lies in​​​‌ a physical prior assumption‌ about the network architecture‌​‌ and its countercurrent flow​​ learning process inspired by​​​‌ the column's operation. Our‌ experimental results showcase a‌​‌ significant improvement in accuracy​​ and highlight the potential​​​‌ of our augmented model‌ for generalizing across domains,‌​‌ especially when data is​​ scarce.

Keywords: Physics-informed ML​​​‌

8.2.7 Image quality metrics‌ for restricted gamut images‌​‌ produced by laser-induced printing​​ on plasmonic films (Journal​​​‌ of the European Optical‌ Society 2025) 9

Participants:‌​‌ Rémi Emonet.

Collaboration​​ with Robin Mermillod-Blondin, Nicolas​​​‌ Dalloz, Alain Tremeau, Aldi‌ Wista Fadhilah, and Nathalie‌​‌ Destouches (LabHC)

Laser-induced printing​​ is a low-cost, high-speed,​​​‌ non-contact method of marking‌ large, highresolution images. Implemented‌​‌ on thin films containing​​ metallic nanoparticles, the technique​​​‌ allows for the printing‌ of color images with‌​‌ visual effects. However, these​​ images typically have a​​​‌ limited color gamut compared‌ to inkjet printing. This‌​‌ limitation is due to​​ the inability to achieve​​​‌ high levels of saturation‌ for all colors and‌​‌ to cover the sRGB​​ hue range. While common​​​‌ quality metrics focus primarily‌ on aspects such as‌​‌ resolution or blur, they​​ rarely address the color​​​‌ aspect. This study proposes‌ a methodology to provide‌​‌ image quality metrics adapted​​ to color gamuts with​​​‌ unusual shapes and volumes.‌ It aims to rank‌​‌ them in terms of​​​‌ image quality performance for​ any given image. In​‌ particular, this work focuses​​ on gamuts measured in​​​‌ transmission and reflection that​ are not necessarily centered​‌ on the CIE a*b*​​ plane and may exhibit​​​‌ low contrast. Psychophysical studies​ have been conducted to​‌ evaluate the quality of​​ images simulated with different​​​‌ color gamuts. The same​ images were evaluated using​‌ different metrics, and an​​ analysis based on the​​​‌ ANOVA model was used​ to determine a set​‌ of metrics that explain​​ observers' preferences.

Keywords: Laser-matter​​​‌ interaction, Machine Learning

Note​ that other works in​‌ physics-informed machine learning have​​ been presented at physics​​​‌ conferences, such as 21​, 26.

8.3.1 Relax and​​​‌ penalize: a new bilevel​ approach to mixed-binary hyperparameter​‌ optimization (TMLR'25) 11

Participants:​​ Jordan Patracone.

Collaboration​​​‌ with Sara Venturini, Marianna​ de Santis, Francesco Rinaldi,​‌ Saverio Salzo (Italy), Martin​​ Schmidt (Germany)

In recent​​​‌ years, bilevel approaches have​ become very popular to​‌ efficiently estimate high-dimensional hyperparameters​​ of machine learning models.​​​‌ However, to date, binary​ parameters are handled by​‌ continuous relaxation and rounding​​ strategies, which could lead​​​‌ to inconsistent solutions. In​ this context, we tackle​‌ the challenging optimization of​​ mixed-binary hyperparameters by resorting​​​‌ to an equivalent continuous​ bilevel reformulation based on​‌ an appropriate penalty term.​​ We propose an algorithmic​​​‌ framework that, under suitable​ assumptions, is guaranteed to​‌ provide mixed-binary solutions. Moreover,​​ the generality of the​​​‌ method allows to safely​ use existing continuous bilevel​‌ solvers within the proposed​​ framework. We evaluate the​​​‌ performance of our approach​ for two specific machine​‌ learning problems, i.e., the​​ estimation of the group-sparsity​​​‌ structure in regression problems​ and the data distillation​‌ problem. The reported results​​ show that our method​​​‌ is competitive with state-of-the-art​ approaches based on relaxation​‌ and rounding.

Keywords: Bilevel​​ Optimization

8.4.1​​ Self-Play Q-Learners Can Provably​​​‌ Collude in the Iterated​ Prisoner's Dilemma (ICML'25) 12​‌

Participants: Quentin Bertrand.​​

Collaboration with Emilio Calvano​​​‌ (LUISS Business School, Università​ LUISS Guido Carli, Rome,​‌ Italy), Juan Agustin Duque​​ and Gauthier Gidel from​​​‌ MILA (Canada)

A growing​ body of computational studies​‌ shows that simple machine​​ learning agents converge to​​​‌ cooperative behaviors in social​ dilemmas, such as collusive​‌ price-setting in oligopoly markets,​​ raising questions about what​​​‌ drives this outcome. In​ this work, we provide​‌ theoretical foundations for this​​ phenomenon in the context​​ of self-play multi-agent Q-learners​​​‌ in the iterated prisoner's‌ dilemma. We characterize broad‌​‌ conditions under which such​​ agents provably learn the​​​‌ cooperative Pavlov (win-stay, lose-shift)‌ policy rather than the‌​‌ Pareto-dominated "always defect" policy.​​ We validate our theoretical​​​‌ results through additional experiments,‌ demonstrating their robustness across‌​‌ a broader class of​​ deep learning algorithms.

Keywords:​​​‌ Q-Learning, Cooperation, Prisoner dilemma‌

9.1.1 Partnership Inria-IFPEN (2025-2028)‌

Participants: Rémi Emonet,‌​‌ Marc Sebban.

Partner​​: IFPEN Solaize -​​​‌ Conception, Modélisation, Procédés.

This‌ newly signed 2025-2028 project‌​‌ is part of the​​ national strategic partnership between​​​‌ Inria and IFPEN in‌ support of the energy‌​‌ transition, which led to​​ the creation of a​​​‌ joint laboratory between the‌ two partners in 2020.‌​‌ It deals with Physics-informed​​ Machine Learning, especially with​​​‌ hybrid deep-learning models that‌ combines physical laws with‌​‌ data-driven approaches. Their promise​​ is to be more​​​‌ interpretable, more robust when‌ facing outliers, and to‌​‌ lead to a more​​ plausible physical behavior than​​​‌ their pure-machine-learning counterparts. The‌ application case is post-combustion‌​‌ carbon capture, a process​​ of major importance to​​​‌ tackle hard-to- abate greenhouse‌ gas emissions of some‌​‌ key industrial sectors. The​​ absorption column of this​​​‌ type of unit is‌ modeled using a rigorous‌​‌ description of the physical​​ mechanisms responsible for the​​​‌ transfer of acid gas‌ to the liquid solvent.‌​‌ Nonetheless, this model fails​​ to predict some experimental​​​‌ tendencies observed on industrial‌ units, and these discrepancies‌​‌ cannot be explained by​​ a specific phenomenon. In​​​‌ this context, the objective‌ of this joint project‌​‌ (including a PhD thesis​​ co-supervised by Rémi Emonet​​​‌ and Marc Sebban )‌ is to design new‌​‌ hybrid PiML approaches, which​​ incorporate laws from chemistry,​​​‌ to improve the precision‌ of the absorption column‌​‌ description.

9.1.2 CIFRE Theses​​ with Thalès (2024-2027)

Participants:​​​‌ Rémi Eyraud.

Partners‌: SESAM (LabHC-UJM), Thalès-DIS‌​‌

During Falls 2024, a​​ collaboration with the DIS​​​‌ (Digital Identity & Security)‌ lab of the Thalès‌​‌ firm started via a​​ CIFRE PhD (Wissal Ghamour).​​​‌ Rémi Eyraud took the‌ academic direction of the‌​‌ PhD together with members​​ of the SESAME team​​​‌ of the Hubert Curien‌ Laboratory. The subject aims‌​‌ at leveraging the tools​​ of Machine Learning for​​​‌ Side Channel Attacks of‌ embedded microchips. Information from‌​‌ physics can be added​​ to the process, since​​​‌ the leak mainly relies‌ on an electro-magnetic sensor.‌​‌ This followed the Ph.D​​ of Dorian Llavata that​​​‌ started in 2022 on‌ the same subject in‌​‌ collaboration with the CESTI​​ Leti of the CEA.​​​‌

9.1.3 I-Démo Région "GREENAI"‌ (2024-2027)

Participants: Jordan Patracone‌​‌.

Partners: Dracula​​ Technologies, ASYGN.

The project​​​‌ I-Démo Région "GREENAI" involves‌ three key actors working‌​‌ collaboratively towards sustainable Internet​​ of Things (IoT) solutions.​​​‌ Dracula Technologies specializes in‌ developing organic photovoltaic cells‌​‌ that harness ambient light​​ to power IoT devices,​​​‌ aligning with environmental sustainability‌ goals. ASYGN contributes by‌​‌ designing ultra-low-power hardware accelerators​​ to enable advanced AI​​​‌ processing directly on IoT‌ devices. The Laboratoire Hubert‌​‌ Curien (Jordan Patracone​​​‌ ) focuses on energy-efficient​ artificial intelligence for computer​‌ vision, optimizing neural network​​ architectures for low-resource hardware​​​‌ through two scientific theses​ (among which, that of​‌ Ben Gao). This work​​ is sponsored by a​​​‌ public grant overseen by​ the Auvergne-Rhône-Alpes region, Grenoble​‌ Alpes Métropole, and BPIFrance.​​

9.2.1 "Baby Cry"​ project - AXA Fundation​‌ (2024-2026)

Participants: Rémi Emonet​​.

Partners: ENES​​​‌ (UJM), SAINBIOSE-MoVE, CHU Saint-Étienne.​

Lead by the ENES,​‌ the AXA Baby Cry​​ 1000 project aims at​​​‌ recording 1000 babies, each​ for one day after​‌ they are born to​​ help early diagnosis of​​​‌ cognitive development issues and​ compare with the development​‌ of premature babies. The​​ project is a joint​​​‌ effort by three labs​ (ENES, Laboratoire Hubert Curien​‌ (Rémi Emonet), and SAINBIOSE-MoVE,​​ CHU Saint-Étienne) and is​​​‌ funded by the AXA​ Fundation for a total​‌ of 1M€ and a​​ duration of 3 years.​​​‌

10.1.1​‌ Inria associate team not​​ involved in an IIL​​​‌ or an international program​

MALICE initiated the creation​‌ of two Inria associate​​ teams in 2025: DYNAMO​​​‌, led by Jordan​ Patracone in collaboration with​‌ IIT (Italy), and​​ LSD, headed by​​​‌ Quentin Bertrand in collaboration​ with MILA (Canada).​‌

10.2.1 Visits‌​‌ of international scientists

10.2.2 Visits‌​‌ to international teams

ML4Health -​​​‌ Transform4Europe (2024-2026)

Participants: Amaury‌ Habrard, Marc Sebban‌​‌.

Partners: University​​ of Alicante, Dpto.Tecnologia Informatica​​​‌ y Computacion (DTIC)

Type‌: Seed Funding T4EU‌​‌ (co-funded by the European​​ Union)

Duration: 2​​​‌ years (2024-2026)

The objective‌ of ML4Health is two-fold:‌​‌ (i) from a training​​ perspective, develop a double​​​‌ master degree in AI‌ between the two T4EU‌​‌ partners, built from two​​ existing complementary programmes (Machine​​​‌ Learning & Data Mining‌ master track at UJM‌​‌ and Máster Universitario en​​ Inteligencia Artificial at UA);​​​‌ (ii) from a scientific‌ standpoint, benefit from the‌​‌ expertise in statistical learning​​ and physics-informed Machine Learning​​​‌ (UJM) and in 3D‌ perception and intelligent systems‌​‌ (DTIC) to focus on​​ the morphological evolution of​​​‌ bodies of obese patients‌ and the early detection‌​‌ of neurodegenerative diseases. In​​ particular, the two partners​​​‌ study (data+theory) hybrid machine‌ learning models combining both‌​‌ 3D+t images and biological/morphometric​​ knowledge.

10.4.1​​ PEPR-IA PRODIGE-AI (2026-2030)

Participants:​​​‌ Amaury Habrard, Quentin‌ Bertrand, Farah Cherfaoui‌​‌, Rémi Emonet,​​ Rémi Eyraud, Benjamin​​​‌ Girault, Jordan Patracone‌, Marc Sebban.‌​‌

PROGIGE-AI: PRObability, ranDom matrIx​​ theory, Geometry and gEneralization​​​‌ for generative-AI

Partners:‌ LIS & IMT (Marseille),‌​‌ LJAD (Nice), LITIS (Rouen),​​ LPT & IMT (Toulouse)​​​‌ & E. Morvant (LabHC-UJM)‌

Type: PEPR-IA project‌​‌

The rapid and impressive​​​‌ development of generative AI​ opens up transformative projects​‌ for several key sectors​​ such as healthcare, education​​​‌ and industry. This evolution​ comes with major challenges​‌ in order to ensure​​ safe, ethical, and effective​​​‌ development of these technologies.​ In particular, ethical issues​‌ such as plagiarism and​​ the amplification of existing​​​‌ biases are currently paramount.​ Guarantees regarding the functioning​‌ of these generative models​​ and the understanding of​​​‌ their underlying mechanisms remain​ preliminary. The PRODIGE-AI project​‌ aims to address three​​ crucial issues for the​​​‌ development of safer, more​ efficient, and more transparent​‌ generative AI: the generalization​​ capabilities of generative AI​​​‌ models, the efficiency and​ explainability of these AI​‌ models, and the development​​ of geometric generative AI​​​‌ - in particular for​ graphs. These three areas​‌ represent topics where the​​ state of the art​​​‌ requires new foundational frameworks​ to enable a controlled​‌ development. Our shared vision​​ is that the promising​​​‌ potential of generative AI​ cannot be realized without​‌ a deep understanding of​​ the fundamentals of generative​​​‌ deep neural networks and​ the mathematics underlying their​‌ learning mechanisms. Towards that​​ endeavor, our consortium brings​​​‌ together both specialists in​ the foundations of machine​‌ learning and mathematicians experts​​ in probability theory, graph​​​‌ theory and geometry. In​ mathematics, the project relies​‌ on expertise in stochastic​​ processes, filtering, high dimensional​​​‌ statistics, random matrices, random​ graphs, random tensors, free​‌ probability, C*-algebras, graph theory,​​ classical and quantum information​​​‌ theory, and optimal transport.​ Another important objective of​‌ the project is to​​ structure a group of​​​‌ researchers at the frontier​ between the foundations of​‌ machine learning and probability​​ theory to foster the​​​‌ development of innovative and​ relevant results for generative​‌ AI.

10.4.2 ANR MONALISA​​ (2025-2029)

MONALISA: Monotone Operators​​​‌ and Neural Architectures -​ Leveraging Interactions for physically​‌ Structured Approximations

Participants: Jordan​​ Patracone.

Partners:​​​‌ CSML / IIT (Italy),​ DIAG / Univ. Sapienza​‌ di Roma (Italy), Univ.​​ Novi Sad (Serbia), ERIC​​​‌ / Univ. Lyon 2​

Type: ANR JCJC​‌

Neural operators have recently​​ shown great success in​​​‌ approximating complex, high-dimensional dynamical​ systems, such as those​‌ governed by partial differential​​ equations (PDEs). Their advantage​​​‌ over traditional neural networks​ lies in their ability​‌ to model mappings between​​ function spaces, capturing continuous​​​‌ dynamics across scales with​ discretization-invariant performance. Physics-informed neural​‌ operators further embed physical​​ constraints, enforcing laws like​​​‌ PDEs. Trained with both​ data-driven and physics-based regularization,​‌ they offer faster predictions​​ than traditional PDE solvers.​​​‌ However, the reasons behind​ their effectiveness remain unclear.​‌ A critical aspect of​​ this ongoing exploration is​​​‌ determining how the complexity​ of these operators is​‌ controlled—whether through explicit regularization​​ techniques or implicitly via​​​‌ their architectural design. In​ MONALISA, we aim to​‌ adopt an ambitious approach​​ and borrow tools from​​​‌ the monotone operator theory,​ an area of nonlinear​‌ analysis recently applied to​​ the study of nonlinear​​​‌ mappings modeled by neural​ architectures, to further understand​‌ and control the complexity​​ of physics-informed neural operators.​​​‌ To implicitly control their​ complexity, the first research​‌ axis will leverage such​​ connection to design neural​​ operators whose architecture itself​​​‌ is guided by physical‌ knowledge such as PDE.‌​‌ A starting point will​​ be to formulate the​​​‌ action of an operator‌ layer as the solution‌​‌ of a regularized optimization​​ problem over function spaces.​​​‌ Then, we will study‌ the impact of explicit‌​‌ regularization both at the​​ layer-level as mentioned above​​​‌ and at the training‌ loss level. Moreover, theoretical‌​‌ guarantees will be devised​​ by hinging on monotone​​​‌ operator theory and on‌ the theory of reproducing‌​‌ kernel Banach spaces. Finally,​​ we will focus on​​​‌ the challenging application of‌ laser-matter interaction, where structuring‌​‌ the neural operator with​​ all available knowledge is​​​‌ essential to overcome the‌ scarcity of data.

10.4.3‌​‌ ANR LSD (2025-2028)

LSD:​​ Leveraging Synthetic Data From​​​‌ Generative Models

Participants: Quentin‌ Bertrand, Rémi Emonet‌​‌, Amaury Habrard,​​ Marc Sebban.

Partners​​​‌: UdeM Mila (Canada),‌ Inria OCKHAM (France)

Type‌​‌: ANR PRCI

Generative​​ models are machine learning​​​‌ models that learn and‌ replicate the underlying structure‌​‌ of the data. The​​ quality of the data​​​‌ they produce has reached‌ a level that surpasses‌​‌ the human ability to​​ differentiate between real and​​​‌ synthetic data. This opens‌ up the possibility of‌​‌ virtually unlimited access to​​ realistic synthetic data, which​​​‌ could be leveraged for‌ data augmentation, especially in‌​‌ situations where real-world data​​ is scarce. In fields​​​‌ such as physics, clinical‌ applications, and protein design,‌​‌ synthetic data can enrich​​ datasets and enhance model​​​‌ generalization.

With the rise‌ of indistinguishable synthetic content‌​‌ being generated and shared​​ online, deployed systems now​​​‌ face the unprecedented challenge‌ of managing synthetic data‌​‌ alongside authentic data. A​​ growing concern in generative​​​‌ AI is "self-consuming" models,‌ which are retrained on‌​‌ their own (previously generated)​​ data. Over time, this​​​‌ recursive process can yield‌ overfitting artifacts, accumulation of‌​‌ biases, or inaccuracies, ultimately​​ causing critical model degradations​​​‌ (a.k.a., model collapse).

Leveraging‌ the complementary expertise of‌​‌ the involved partners in​​ generative modelling, optimization, and​​​‌ open-source software, the project‌ aims to systematically investigate‌​‌ the risks and benefits​​ of interactions between learning​​​‌ algorithms and synthetic data.‌ One key objective is‌​‌ to assess when and​​ to what extent generative​​​‌ models can improve performance‌ on downstream tasks. Another‌​‌ goal is to quantify​​ the rate at which​​​‌ "self-consuming" models collapse and‌ to develop strategies to‌​‌ mitigate it. More broadly,​​ the project will explore​​​‌ how generative models behave‌ and interact when deployed‌​‌ in shared environments with​​ multiple models or agents.​​​‌ This includes understanding how‌ they influence each other’s‌​‌ outputs, how they potentially​​ cooperate or compete, and​​​‌ the impact of these‌ interactions on overall system‌​‌ performance.

10.4.4 ANR MELISSA​​ (2024-2029)

MELISSA: MEthodological contributions​​​‌ in statistical Learning InSpired‌ by SurfAce engineering

Participants:‌​‌ Eduardo Brandao, Rémi​​ Emonet, Benjamin Girault​​​‌, Amaury Habrard,‌ Jordan Patracone, Marc‌​‌ Sebban.

Partners:​​ MAGNET (Inria Lille), MLIA​​​‌ (ISIR, PSL) and F.‌ Garrelie and JP Colombier‌​‌ (LabHC-UJM)

Type: ANR​​ PRC

Website: MELISSA​​​‌

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. Despite indisputable​​​‌ advances, several open problems​ remain to be addressed​‌ in PIML: (i) Deriving​​ generalization guarantees; (ii) Learning​​​‌ with a limited amount​ of data; (iii) Augmenting​‌ partially known physical laws;​​ (v) Modeling uncertainty; (vi)​​​‌ Building foundation models for​ physics. MELISSA will deal​‌ with these problems 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.​

10.4.5 ANR DATES (2025-2029)​‌

DATES: Domain Adaptation on​​ Time Series

Participants: Rémi​​​‌ Emonet, Farah Cherfaoui​, Quentin Bertrand,​‌ Marc Sebban.

Partners​​: IRISA (Rennes), ERICSSON​​​‌ (Firm)

Type: ANR​ PRCE

Machine learning models​‌ generally assume that training​​ and test data are​​​‌ identically distributed. However, this​ assumption often breaks down​‌ in real-world applications due​​ to shifting data distributions​​​‌ over time and across​ contexts. Domain Adaptation (DA)​‌ tackles this challenge by​​ allowing models to leverage​​​‌ labeled data from one​ domain to perform tasks​‌ in a different, often​​ unlabeled, target domain. Time​​​‌ Series (TS) consists of​ data points recorded over​‌ time and is increasingly​​ prevalent with autonomous sensors​​​‌ and online activities. In​ the context of DA,​‌ TS presents several challenges,​​ including class imbalance, temporal​​​‌ drift, correlations and dependencies,​ and hierarchical structures that​‌ manifest as multi-scale periodic​​ patterns. The DATeS project​​​‌ addresses three fundamental challenges​ of DA for TS.​‌ The first challenge is​​ efficiently achieving both temporal​​​‌ and domain alignment, which​ we will tackle by​‌ proposing new formulations based​​ on generalized optimal transport​​​‌ and conditional flow matching.​ The second challenge involves​‌ adapting to missing data​​ and source-free domain adaptation,​​​‌ where only a learned​ model is available; we​‌ will tackle this in​​ the lens of optimal​​​‌ transport approaches. The third​ challenge relates to unsupervised​‌ DA (with few or​​ no target labels) and​​​‌ involves the general problem​ of selecting the best​‌ model, source domain(s), and​​ hyperparameters without access to​​​‌ a validation set. We​ will address this challenge​‌ by learning metrics tailored​​ for TS in DA​​​‌ and through meta-learning, enabling​ effective adaptation from multiple​‌ sources. This project aims​​ to generate significant new​​​‌ knowledge in the field​ of DA for TS,​‌ addressing critical challenges that​​ are currently under-explored. While​​​‌ DATeS is motivated by​ the challenges posed by​‌ time series data, it​​ will also inspire innovative​​​‌ developments that have broader​ applicability beyond the TS​‌ domain.

10.4.6 ANR ACOULAK​​ (2025-2029)

ACOULAK: Underwater acoustic​​​‌ pollution in peri-alpine lakes:​ Characterization and ecological impacts​‌

Participants: Rémi Emonet,​​ Marc Sebban.

Partners​​​‌: ENES (UJM, Coordination),​ CARRTEL, INRA

Type:​‌ ANR PRC

Lakes are​​ key resources for territories,​​ providing many ecosystem services​​​‌ like water supply, aquatic‌ biodiversity and human wellbeing.‌​‌ With the rise of​​ recreational activities, lake managers​​​‌ worry about the increase‌ in human presence on‌​‌ lakes with ever more​​ powerful boats. One of​​​‌ the symptoms of water‌ overcrowding is acoustic pollution.‌​‌ Boat sounds propagate in​​ the air and generate​​​‌ conflicts between the populations‌ of users, but also‌​‌ underwater where they are​​ likely to disturb the​​​‌ aquatic fauna. Concern for‌ freshwater acoustic pollution is‌​‌ very recent and ecosystem​​ managers are seeking for​​​‌ assessment tools and fundamental‌ knowledge on the ecological‌​‌ impacts to develop sustainable​​ management strategies that promote​​​‌ human activities while preserving‌ ecosystem integrity. We propose‌​‌ a timely and innovative​​ project on underwater acoustic​​​‌ pollution using the peri-alpine‌ lakes as model systems.‌​‌ Our first objective is​​ to characterize underwater acoustic​​​‌ pollution in space and‌ time using passive acoustic‌​‌ monitoring as new metrological​​ tool and machine learning​​​‌ to process the data.‌ We hypothesize that acoustic‌​‌ pollution varies in time​​ depending on human activities​​​‌ and climatic variables, and‌ in space according to‌​‌ local practices and regulations.​​ Our second objective is​​​‌ to investigate impacts on‌ dreissenid mussels and Eurasian‌​‌ perch larvae using playback​​ experiments in the lab​​​‌ and the field. We‌ hypothesize that increased sound‌​‌ level induces physiological and​​ behavioral symptoms of a​​​‌ stress response. The consortium‌ brings together experts from‌​‌ the fields of ecology,​​ animal behavior, limnology and​​​‌ computer sciences. We will‌ provide fundamental knowledge with‌​‌ international publications in the​​ fields of environmental pollution,​​​‌ animal ecology, and machine‌ learning, but also a‌​‌ unique database on freshwater​​ underwater soundscapes, decision supports​​​‌ for lake managers, and‌ materials for public awareness‌​‌ events.

10.4.7 ANR TAUDoS​​ (2021-2026)

Participants: Rémi Emonet​​​‌, Rémi Eyraud,‌ Amaury Habrard, Marc‌​‌ Sebban.

TAUDoS: Theory​​ and Algorithms for the​​​‌ Understanding of Deep learning‌ On Sequential data

Partners‌​‌: LIS (Aix-Marseille University),​​ EURA NOVA (Firm), MILA​​​‌ (Canadian State)

Type:‌ ANR PRCE

Website:‌​‌ TAUDoS

The ambition of​​ this project is to​​​‌ provide a better understanding‌ of the mechanisms that‌​‌ allow the amazing recent​​ achievements of Machine Learning,​​​‌ and in particular of‌ Deep Learning. This is‌​‌ achieved by providing elements​​ that allow a better​​​‌ scientific comprehension of the‌ models, strengthening our experimental‌​‌ results by theoretical guarantees,​​ incorporating components dedicated to​​​‌ interpretability within the models,‌ and allowing trustful quantitative‌​‌ comparison between learned models.​​

The originality and the​​​‌ specificities of TAUDoS are‌ due to three major‌​‌ characteristics:

• The focus on​​ models for sequential data,​​​‌ such as Recurrent Neural‌ Networks (RNN), while most‌​‌ works concentrate on feed-forward​​ networks
• The will to​​​‌ analyze these models in‌ the light of formal‌​‌ language theory
• The goal​​ to target both rigorous​​​‌ theoretical analyses and empirical‌ evidence related to interpretability.‌​‌

10.4.8 ANR SAFE (2022-2026)​​

SAFE: Controlling networks with​​​‌ safety bounded and interpretable‌ machine learning

Participants: Amaury‌​‌ Habrard.

Partners:​​ XLIM / Univ. Poitiers,​​​‌ IRISA / Univ. Rennes‌ 1, Huawei, QOS DESIGN‌​‌ and B. Jeudy &​​​‌ K. Singh (LabHC-UJM)

Type​: ANR PRCE

When​‌ applied to communication networks,​​ traditional approaches for control​​​‌ and decision-making require a​ comprehensive knowledge of system​‌ and user behaviours, which​​ is unrealistic in practice​​​‌ when there is an​ increase in scale and​‌ complexity. Data-driven AI approaches​​ do not have this​​​‌ drawback, but offer no​ safety bounds and are​‌ difficult to interpret. The​​ SAFE project aims to​​​‌ design an innovative approach​ by combining the best​‌ of both worlds. In​​ this new approach, intelligence​​​‌ is distributed in the​ network between a global​‌ AI (at the central​​ level) and local AIs​​​‌ (at the edge level)​ collaborating with each other​‌ by integrating traditional models​​ with graph neural networks​​​‌ and reinforcement learning. The​ approach, developed for partially​‌ or completely observable/controllable environments,​​ will natively integrate safety​​​‌ bounds, interpretability and provide​ self-adaptive systems for routing,​‌ traffic engineering and scheduling.​​ SAFE has following scientific​​​‌ objectives with an open​ source strategy:

1. Hierarchical architecture:​‌ Assuming modern network architectures,​​ we will design a​​​‌ ML architecture based on​ global AI (running at​‌ central controller level) and​​ local AI (running at​​​‌ edge device level) for​ decision-making in partially as​‌ well as fully observable​​ and controllable environments. Global​​​‌ AI will be able​ to control, configure and​‌ install policies on local​​ AI.
2. Algorithms for partially​​​‌ observable environments: We will​ design new safety bounded​‌ and interpretable algorithms for​​ self-adaptive traffic engineering, automatic​​​‌ scheduling algorithms for partially​ observable and controllable environments.​‌ These methods find use​​ cases in SD-WAN (Software-Defined​​​‌ Wide Area Networks), where​ edge devices present at​‌ customer premises need to​​ collaboratively operate in overlay​​​‌ on top of partially​ observable core networks.
3. Algorithms​‌ for fully observable environments:​​ We will investigate the​​​‌ application of the global​ and local AI architecture​‌ for fully observable and​​ controllable environments. Specifically, we​​​‌ will design new safety​ bounded and interpretable algorithms​‌ for software-defined routing and​​ traffic engineering, which find​​​‌ use cases in data​ centers as well as​‌ private WANs connecting multiple​​ sites.

10.4.9 ANR FAMOUS​​​‌ (2023-2027)

Participants: Rémi Eyraud​, Amaury Habrard.​‌

Famous: Fair Multi-modal Learning​​

Partners: LIS Aix-Marseille,​​​‌ LITIS Rouen, INT Marseille,​ Euranova and A. Gouru,​‌ C. Largeron & E.​​ Morvant (LabHC-UJM)

The aim​​​‌ of this project is​ to explore the first​‌ avenues of research into​​ the contribution of multimodality​​​‌ in datasets to meet​ the requirements of fair​‌ learning. Fairness refers here​​ to the biases (in​​​‌ the data and/or induced),​ while being interested in​‌ the interpretability of the​​ models to help their​​​‌ certification. Each modality has​ its own statistical and​‌ topological characteristics, which requires​​ upstream research on the​​​‌ adjustment of distributions when​ biased, adapted metrics, etc.​‌ Moreover, each one being​​ itself a bias of​​​‌ observation of the data,​ this will be taken​‌ into account to establish​​ a joint distribution (trans-modal)​​​‌ unbiased on all these​ modalities. With theoretical research​‌ in cross-modal statistical learning,​​ we will study methods​​​‌ for reducing some types​ of identified biases (non​‌ iid, imbalances, sensitive variables)​​ in the case of​​ multimodal data. Two levels​​​‌ of treatment are privileged:‌ (1) cross-modal pre-processing of‌​‌ biases in the data,​​ by learning metrics, neural​​​‌ representations, and optimization constraints‌ on kernel pre-images; (2)‌​‌ cross-modal algorithms for eliminating​​ biases in model learning:​​​‌ cross-modal optimization algorithms, as‌ well as optimal transfer‌​‌ and transport approaches between​​ modalities to debias the​​​‌ concerned ones, based on‌ the theoretical results previously‌​‌ obtained. Parsimony will be​​ considered for scaling and​​​‌ explainability. Transversally, our work‌ will be based on‌​‌ problems arising from real​​ data sets in biology​​​‌ and health, multi-modal and‌ presenting various types of‌​‌ bias, and on toy​​ data sets to be​​​‌ generated. They have modalities‌ where the data are‌​‌ structured in graphs: all​​ our fundamental works will​​​‌ be declined to take‌ into account this specificity‌​‌ impacting the treatment of​​ the considered biases.

10.4.10​​​‌ EUR SLEIGHT TREASURF (2024-2027)‌

TREASURF: TRansfer lEarning for‌​‌ Frugal and Accurate modeling​​ of SURface Functionalization prediction​​​‌ –application to multicomponent alloys‌

Participants: Amaury Habrard,‌​‌ Rémi Emonet, Marc​​ Sebban.

Partners:​​​‌ F. Garrelie and JP‌ Colombier (LabHC-UJM)

Type:‌​‌ EUR Manutech SLEIGHT

TREASURF​​ is an interdisciplinary project​​​‌ focusing on the development‌ of novel machine learning‌​‌ approaches for the prediction​​ of surface functionalization of​​​‌ different families of metals‌ and metal alloys by‌​‌ (femto)laser irradiation. The ability​​ to predict the micro-​​​‌ or nanopatterns induced by‌ laser functionalization is a‌​‌ crucial challenge for an​​ optimal use of surface​​​‌ properties. In this context,‌ machine learning methods have‌​‌ been subject of a​​ growing interest recently but​​​‌ they have to cope‌ with of limited amounts‌​‌ of experimental data due​​ to the very high​​​‌ acquisition costs. In the‌ TREASURF project, we propose‌​‌ to address this problem​​ by developing methods able​​​‌ to transfer the knowledge‌ of a prediction model‌​‌ learned from a given​​ metal or alloy to​​​‌ another, different but sharing‌ certain properties. Re-training a‌​‌ new model is not​​ a plausible hypothesis, mainly​​​‌ because of the difficulties‌ involved in acquiring large‌​‌ quantities of data (laser​​ irradiation + nanoscale imaging).​​​‌ This project is therefore‌ situated in a difficult‌​‌ context of "frugal" learning.​​ Our aim is to​​​‌ focus primarily on topographic‌ predictions for two or‌​‌ more different alloy families.​​ The project also envisages​​​‌ taking into account variability‌ due to chemical changes‌​‌ to guide the transfer​​ process. The advances made​​​‌ in this project will‌ enable us to better‌​‌ characterize the impact of​​ laser-matter interaction with the​​​‌ perspective of designing new‌ surface functionalizations on various‌​‌ novel metal alloys, opening​​ the door to new​​​‌ application prospects in numerous‌ societal challenges related to‌​‌ health, energy, space, nuclear​​ or defense.

10.4.11 EUR​​​‌ SLEIGHT PROXIMA (2025-2027)

PROXIMA:‌ Unfolded proximal neural operators‌​‌ for inverse problems in​​ imaging

Participants: Jordan Patracone​​​‌.

Partners: L.‌ Denis (LabHC-UJM)

Type:‌​‌ EUR Manutech SLEIGHT

This​​ project will develop a​​​‌ novel framework for solving‌ inverse problems in imaging,‌​‌ focusing on the problem​​ of joint reconstruction of​​​‌ a sequence of high-quality‌ images from multiple degraded‌​‌ or incomplete observations. Classical​​​‌ approaches often rely on​ discretized formulations that are​‌ sensitive to resolution changes,​​ leading to poor generalization​​​‌ across different scales. By​ operating directly in function​‌ spaces, our approach will​​ ensure discretization invariance, allowing​​​‌ models trained on one​ resolution to generalize seamlessly​‌ to others. In this​​ context, we will define​​​‌ and explore the extension​ of unfolded proximal networks​‌ – which design the​​ deep network architecture itself​​​‌ by leveraging the variational​ formulation associated with the​‌ inverse problem – to​​ function spaces, leading to​​​‌ unfolded proximal neural operators.​ This will allow for​‌ an adaptive and principled​​ way to encode physical​​​‌ constraints and optimization-based priors​ directly into the learning​‌ process. Such an approach​​ is particularly well-suited for​​​‌ ill-posed settings and applications​ where the sample undergoes​‌ unknown deformations, where classical​​ data-driven methods often fail.​​​‌ A key challenge will​ be to extend generalization​‌ guarantees of unfolded networks​​ beyond discrete settings, requiring​​​‌ new theoretical developments at​ the intersection of operator​‌ learning, proximal optimization, inverse​​ problems, and statistical learning​​​‌ theory. We will also​ design self-supervised training strategies​‌ so that our methods​​ can be applied to​​​‌ experimental data without requiring​ ground-truth images during the​‌ learning step (i.e., the​​ clean and noise-free images​​​‌ corresponding to each actual​ observation). This framework will​‌ offer a scalable and​​ interpretable solution for high-dimensional​​​‌ inverse problems in imaging,​ with potential applications in​‌ surface characterization or biological​​ sample reconstruction.

11.1.1 Scientific events: organisation​‌

11.1.2 Scientific events: selection​

11.1.3 Journal

11.1.4 Invited‌ talks

• Quentin Bertrand :‌​‌ "Some Challenges Around​​ Retraining Generative Models on​​​‌ their Own Data",‌ Imaging in Paris, January‌​‌ 2025.
• Quentin Bertrand ,​​ Rémi Emonet : "Generative​​​‌ Models Tutorial", SenHubAI Summer‌ School, April 2025.
• Quentin‌​‌ Bertrand : "On​​ the Closed Form of​​​‌ Flow Matching: Generalization Does‌ Arise From Target Stochasticity‌​‌", NeurIPS, December 2025.​​
• Rémi Emonet : "​​​‌On the Closed Form‌ of Flow Matching: Generalization‌​‌ Does Arise From Target​​ Stochasticity", EurIPS, December​​​‌ 2025.
• Rémi Emonet "‌Evolution of Generative Models:‌​‌ Diving into CFM",​​ IRISA, Rennes, February 27,​​​‌ 2025,
• Rémi Emonet "‌Generalizing and Scaling Optimal‌​‌ Transport", 10 ans​​ de la FIL, Lyon,​​​‌ May 26, 2025.
• Rémi‌ Emonet "Generative Models:‌​‌ Diving into CFM",​​ SenHubIA Spring School, Remote,​​​‌ March 11, 2025.
• Rémi‌ Emonet "Understanding Generalization‌​‌ In Conditional Flow Matching​​", SMAI Minisymposium «​​​‌ Modèles génératifs, OT et‌ restauration d'images », Carcans,‌​‌ June 3, 2025.
• Rémi​​ Emonet "Creativity in​​​‌ Generative AI", Design,‌ interfaces and digital perspectives,‌​‌ Saint-Étienne, May 22, 2025.​​
• Amaury Habrard : "​​​‌An introduction to Physics-Informed‌ Machine Learning and Applications‌​‌", invited speaker, NormaSTIC​​ Day, April 1, 2025.​​​‌
• Amaury Habrard : "‌What to do with‌​‌ AI in the field​​ of creation—a vision of​​​‌ the limits". Research‌ Day ECCLA-Esadse-Ensba. January 29,‌​‌ 2025.
• Marc Sebban :​​ "PDE discovery from​​​‌ scarce data: a key‌ challenge in physics-informed machine‌​‌ learning", Univ. Alicante,​​ Spain, June 29, 2025.​​​‌
• Eduardo Brandao "Measuring‌ the Dynamic Complexity of‌​‌ Self-Organized Systems", SCCS​​ 2025, May 8, 2025,​​​‌ Istanbul.
• Eduardo Brandao "‌A measure of dynamic‌​‌ complexity of self-organized systems​​", CCS/FR 2025, June​​​‌ 23, 2025, Paris.

11.1.5‌ Scientific expertise

• Marc Sebban‌​‌ , scientific expert for​​ the "Dispositif Doctorants" Normandy​​​‌ Region, 2025

11.1.6 Research‌ administration

• Rémi Emonet is‌​‌ head of the Machine​​ Learning project at laboratoire​​​‌ Hubert Curien.
• Amaury Habrard‌ is head of the‌​‌ Data Intelligence team, member​​ of the advisory committee​​​‌ of the Artificial Intelligence‌ Lyon Saint-Etienne consortium AILyS,‌​‌ member of the scientific​​ board of the MIAI​​​‌ AI Cluster, co-head of‌ the AI and ML‌​‌ axis of the Fédération​​ des Laboratoires d'Informatique de​​​‌ Lyon (FIL), co-responsible of‌ the working group on‌​‌ Physics-aware Machine Learning of​​ the GdR IASIS
• Marc​​​‌ Sebban is deputy director‌ of the Hubert Curien‌​‌ Lab (UMR CNRS 5516),​​ Member of the Board​​​‌ of Directors of the‌ Fédération des Laboratoires d'informatique‌​‌ de Lyon (FIL), Member​​ of the executive committee​​​‌ of the EUR Manutech‌ Sleight, Member of the‌​‌ COMEX Labex MILYON, Member​​​‌ of the COS Inria​ Lyon Centre.

11.2.1 Teaching

• Quentin​ Bertrand :
  • Computer Science​‌ Master, Deep Generative Models,​​ 10h, M2, ENS Lyon.​​​‌
  • Computer Science Master, Numerical​ Optimal Transport for Machine​‌ and Deep Learning, 10h,​​ M2, ENS Lyon.
• Eduardo​​​‌ Brandao :
  • Diplôme Universitaire​ Cycle initial en Technologie​‌ de l’Information de Saint-Étienne:​​ Mathèmatiques pour l'ingénieur 1,​​​‌ 48h, L1, Télécom Saint-Étienne,​ UJM
  • Diplôme Universitaire Cycle​‌ initial en Technologie de​​ l’Information de Saint-Étienne: Mathèmatiques​​​‌ pour l'ingénieur 2, 49.5h,​ L1, Télécom Saint-Étienne, UJM​‌
  • Formation Ingénieur en Apprentissage​​ Data Engineering: Big Data​​​‌ Project, 11h, 3A, Télécom​ Saint- Étienne, UJM
• Farah​‌ Cherfaoui :
  • Licence Mathématiques-Physique-Chimie:​​ Outils Informatiques, 83h, L1,​​​‌ Faculté des Sciences, UJM.​
  • Licence informatique: Programmation Impérative,​‌ 72h, L1, Faculté des​​ Sciences, UJM.
  • Master Machine​​​‌ Learning and Data Mining:​ Advanced Machine Learning, 4h,​‌ M2, Faculté des Sciences,​​ UJM.
  • Licence 3 informatique:​​​‌ Probabilités-Statistiques, 24h, L3, Faculté​ des Sciences, UJM.
  • Licence​‌ 3 informatique: Langage formel​​ 2, 36h, L3, Faculté​​​‌ des Sciences, UJM.
• Rémi​ Emonet :
  • Master Machine​‌ Learning and Data Mining:​​ Probablistic Graphical Models, 20h,​​​‌ M2, Faculté des Sciences,​ UJM.
  • Master Données et​‌ Systèmes Connectés: Programmation Web​​ Avancée, 30h, M1, Faculté​​​‌ des Sciences, UJM.
• Rémi​ Eyraud :
  • Master Machine​‌ Learning and Data Mining:​​ Research Methodology, 20h, 1A,​​​‌ Faculté des Sciences, UJM.​
  • Master Données et Systèmes​‌ Connectés: Research Methodology, 20h,​​ 2A, Faculté des Sciences,​​​‌ UJM.
  • Master Machine Learning​ and Data Mining: Machine​‌ Learning Fundamentals and Algorithms,​​ 30h, 1A, Faculté des​​​‌ Sciences, UJM.
  • Master Données​ et Systèmes Connectés: Machine​‌ Learning Fundamentals and Algorithms,​​ 30h, 1A, Faculté des​​​‌ Sciences, UJM.
  • Master Machine​ Learning and Data Mining:​‌ Advanced Machine Learning, 3h,​​ M2, Faculté des Sciences,​​​‌ UJM.
  • GACO: Base de​ Données, 18h, 2A, IUT​‌ de Saint-Etienne, UJM.
  • GACO:​​ Conception de Sites Web​​​‌ Dynamiques, 112h, 2A, IUT​ de Saint-Etienne, UJM.
  • GACO:​‌ Etablir le Diagnostique Marketing​​ d'une Organisation (SAE), 8h,​​​‌ 2A, IUT de Saint-Etienne,​ UJM.
  • GACO: Traitement des​‌ Données, 20h, 3A, IUT​​ de Saint-Etienne, UJM.
  • GACO:​​​‌ Environnement Informatique, 16h, 1A,​ IUT de Saint-Etienne, UJM.​‌
• Benjamin Girault :
  • Master​​ Machine Learning and Data​​​‌ Mining: Deep Learning I,​ 20h, M1, Faculté des​‌ Sciences, UJM.
• Amaury Habrard​​ :
  • Master Machine Learning​​​‌ and Data Mining: Advanced​ Algorithms and Programming, 20h,​‌ M2, Faculté des Sciences,​​ UJM.
  • Master Machine Learning​​​‌ and Data Mining: Advanced​ Machine Learning, 20h, M2,​‌ Faculté des Sciences, UJM.​​
  • joint course Master Machine​​​‌ Learning and Data Mining​ and Master Données and​‌ Systèmes Connectés: Deep Learning​​ II, 15h, M2, Faculté​​​‌ des Sciences, UJM.
• Jordan​ Patracone :
  • Formation Ingénieur​‌ Télécom Saint-Étienne: Langage C​​ Algorithmie et structures de​​​‌ données, 13.5h, 1A, Telecom​ Saint-Etienne, UJM.
  • Formation Ingénieur​‌ Télécom Saint-Étienne: Algorithmique et​​ structures de données, 4.5h,​​​‌ 1A, Telecom Saint-Etienne, UJM.​
  • Formation Ingénieur Télécom Saint-Étienne:​‌ Projet recherche et innovation,​​ 5h, 3A, Telecom Saint-Etienne,​​​‌ UJM.
  • Formation Ingénieur Télécom​ Saint-Étienne: Big Data Project,​‌ 10h, 3A, Telecom Saint-Etienne,​​ UJM.
  • Formation Ingénieur Télécom​​​‌ Saint-Étienne: Projet d'ingéniérie, 14h,​ 2A, Telecom Saint-Etienne, UJM.​‌
  • Formation Ingénieur Télécom Saint-Étienne:​​ Python et sciences des​​ données, 18h, 2A, Telecom​​​‌ Saint-Etienne, UJM.
  • Formation Ingénieur‌ Télécom Saint-Étienne: Machine learning,‌​‌ 37.5h, 2A, Telecom Saint-Etienne,​​ UJM.
  • Formation Ingénieur en​​​‌ Apprentissage Data Engineering: Statistiques‌ inférentielles, 33h, 2A, Telecom‌​‌ Saint-Etienne, UJM.
  • Formation Ingénieur​​ en Apprentissage Data Engineering:​​​‌ Algorithms for data analysis,‌ 14h, 2A, Telecom Saint-Etienne,‌​‌ UJM.
  • Formation Ingénieur en​​ Apprentissage Data Engineering: Qualité​​​‌ de fonctionnement, 11h, 3A,‌ Telecom Saint-Etienne, UJM.
  • Formation‌​‌ Ingénieur en Apprentissage Data​​ Engineering: Innovation en data,​​​‌ 2h, 3A, Telecom Saint-Etienne,‌ UJM.
• Marc Sebban :‌​‌
  • Licence Informatique: Probabilités-Statistiques, 24h,​​ L3, Faculté des Sciences,​​​‌ UJM.
  • Master Machine Learning‌ and Data Mining: Introduction‌​‌ to Machine Learning, 20h,​​ M1, Faculté des Sciences,​​​‌ UJM.
  • Master Machine Learning‌ and Data Mining: Data‌​‌ Analysis, 24h, M1, Faculté​​ des Sciences, UJM.
  • Master​​​‌ Machine Learning and Data‌ Mining: Advanced Machine Learning,‌​‌ 12h, M2, Faculté des​​ Sciences, UJM.
  • Master Machine​​​‌ Learning and Data Mining:‌ Machine Learning Project, 20h,‌​‌ M2, Faculté des Sciences,​​ UJM.

11.2.2 Supervision

• PhD​​​‌ in progress: Michael Gault.‌ Développement et évaluation d’approches‌​‌ hybrides de machine learning​​ informé par la physique​​​‌ pour la modélisation d’une‌ unité de captage de‌​‌ CO2, since Octo. 2025.​​ Marc Sebban and Rémi​​​‌ Emonet .
• PhD in‌ progress: Petros Kafkas. Harmonic‌​‌ Neural Networks for Physics-informed​​ Machine Learning, since Octo.​​​‌ 2025. Marc Sebban and‌ Benjamin Girault .
• PhD‌​‌ in progress: Diego Pinto-Suárez,​​ Monotone Operators and Neural​​​‌ Architectures: Leveraging Interactions for‌ physically Structured Approximations, since‌​‌ Dec. 2025. Jordan Patracone​​ , Marc Sebban .​​​‌
• PhD defended in 12/2025:‌ Fayad Ali Banna. From‌​‌ Energy Absorption to Self-Organization:​​ A Physics-Informed Machine Learning​​​‌ Approach to Laser-Matter Interaction.‌ Marc Sebban and Rémi‌​‌ Emonet .
• PhD in​​ progress: Hind Atbir. Learning​​​‌ fair and robust kernel-based‌ models with generalization guarantees,‌​‌ since Oct. 2024. Farah​​ Cherfaoui .
• Postdoc in​​​‌ progress: Antoine Caradot. On‌ the estimation of integrals‌​‌ of functions: applications in​​ PIML, since Oct. 2024.​​​‌ Rémi Emonet , Amaury‌ Habrard and Marc Sebban‌​‌ .
• PhD in progress:​​ Wissal Ghamour. Nouvelles approches​​​‌ de l’Intelligence Artificielle pour‌ les attaques par canaux‌​‌ auxiliaires, since Dec. 2024.​​ Rémi Eyraud .
• PhD​​​‌ defended in 12/2025: Dorian‌ Llavata. Apprentissage Profond Diversement‌​‌ Supervisé Pour Les Attaques​​ Par Canaux Auxiliaires, Rémi​​​‌ Eyraud .
• PhD in‌ progress: Robin Mermillod-blondin, Optimisation‌​‌ multicritère de couleurs plasmoniques​​ pour les documents d’identité.​​​‌ Rémi Emonet .
• PhD‌ defended in 11/2025: Sayan‌​‌ Chaki, Unsupervised Analysis of​​ Ornaments extracted from Ancient​​​‌ Books, Rémi Emonet .‌
• PhD in progress: Abdel-Rahim‌​‌ Mezidi, Unveiling and Incorporating​​ Knowledge in Physics-Guided Machine​​​‌ Learning Models, Amaury Habrard‌ , Jordan Patracone .‌​‌
• PhD in progress: Ben​​ Gao, Toward frugal machine​​​‌ learning with physics-aware models,‌ Jordan Patracone .
• PhD‌​‌ in progress: Thibault Girardin,​​ Prediction of multidimensional colors​​​‌ printed by laser on‌ plasmonic metamaterials using deep‌​‌ learning and adaptive strategies,​​ Amaury Habrard .
• PhD​​​‌ in progress: Erick Gomez,‌ TRansfer lEarning for Frugal‌​‌ and Accurate modeling of​​ SURface Functionalization prediction –​​​‌ application to multicomponent alloys,‌ Amaury Habrard .
• PhD‌​‌ Defended in 01/2025: Volodimir​​ Mitarchuk, Theory and Algorithms​​​‌ for the Understanding of‌ Deep Neural Network on‌​‌ Sequential Data, Rémi Eyraud​​​‌ , Amaury Habrard ,​ Rémi Emonet . adaptive​‌ strategies.
• Postdoc (Jan-Oct 2025):​​ Volodimir Mitarchuk, Operator learning​​​‌ in reproducing kernel Banach​ spaces, Jordan Patracone .​‌

11.2.3 Juries

• Marc Sebban​​ : Fabien Lionti (PhD,​​​‌ Univ. Côte d'Azur, reviewer),​ Alice Lacan (PhD, Univ.​‌ Paris-Saclay, President), Lawrence Stewart​​ (PhD, Univ. Paris-Saclay, President).​​​‌
• Rémi Emonet : Mehran​ Adibi (PhD, Université de​‌ Lorraine, Reviewer), Fayad Ali​​ Banna (PhD, Univ. Saint-Etienne,​​​‌ Supervisor), Sayan Chaki (PhD,​ Univ. Saint-Etienne, Supervisor).
• Rémi​‌ Eyraud : Chandrasekar Subamani​​ Narayana (PhD, Aix-Marseille Univ.,​​​‌ reviewer), Volodimir Mitarchuck (PhD,​ Univ. Saint-Etienne, Director)
• Amaury​‌ Habrard : Myriam Tami​​ (HDR, Centrale Supelec-Univ. Paris-Saclay,​​​‌ Reviewer), Marc Lafon (PhD,​ CNAM - Sorbonne Univ,​‌ Reviewer), Jean-Rémy Conti (PhD,​​ Telecom Paris, Reviewer), Rui​​​‌ Yang (PhD, Ecole Centrale​ de Lyon, Member), Louis​‌ Serrano (PhD, Sorbonne Univ.,​​ Reviewer) Lucas Gnecco-Heredia (PhD,​​​‌ Univ Paris-Dauphine PSL, member),​ Kanishka Gosh Dastidar (PhD,​‌ Univ. Passau, Germany, Reviewer),​​ Volidimir Mitarchuk (PhD, Univ.​​​‌ Saint-Etienne, Co-director)

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

• Rémi Emonet, Quentin Bertrand,​​​‌ collaboration with Ockham, we​ wrote a friendly blog​‌ post on recent flow​​ matching techniques, April 2025.​​​‌
• Rémi Eyraud : filmed​ interview at the radio​‌ Ici Saint-Etienne during the​​ AI international submit in​​​‌ Paris, 10/02/2025.
• Quentin Bertrand​ : our recent works​‌ on self-consuming generative models​​ and their biases got​​​‌ media coverage from Sciences​ et Avenir! 09/2025.​‌

11.3.2 Participation in Live​​ events

• Rémi Eyraud :​​​‌ 45 min of conference​ titled "Qu'est-ce que l'IA?",​‌ festival du cinéma jeune​​ public Tête de mule​​​‌, Cinéma Méliès de​ Saint-Etienne, 200 kids, 27/04/2025.​‌
• Rémi Eyraud : 2h​​ of conference titled "de​​​‌ l’Algorithme à l’IA". Complètement​ conf' – Mairie de​‌ Saint-Genest-Lerpt, 50 persons, 13/05/2025.​​
• Amaury Habrard and Rémi​​​‌ Eyraud : an evening​ of science popularization titled​‌ "Vous ne direz plus​​ Intelligence Artificielle !",​​​‌ festival Pint of Science,​ pub 6 Nations, Saint-Etienne,​‌ 50 persons, 21/05/2025
• Rémi​​ Eyraud : 2h of​​​‌ conference and workshops titled​ "Du dieu Algorithme à​‌ la déesse IA", Service​​ Universitaire de Pédagogie, IUT​​​‌ de Saint-Etienne, 28/05/2025.
• Rémi​ Eyraud : 2h of​‌ conference and workshops titled​​ "L'IA et les Associations​​​‌ Familiales", Union des Associations​ Familiales de la Loire,​‌ Chateau de Magneux-Haute-Rive, 70​​ persons, 10/10/2025.
• Rémi Eyraud​​​‌ : 2h of conference​ titled "Démystifier l’IA". L'escale,​‌ Saint-Martin-en-Haut, 3/11/2025.
• Rémi Eyraud​​ : 1h30 of conference​​​‌ titled "Liberté, Algorithmes et​ Responsabilités face à l'IA",​‌ semaine de la laïcité,​​ Médiathèque de Saint-Etienne, 13/12/2025.​​​‌

11.3.3 Others science outreach​ relevant activities

• Amaury Habrard​‌ : round table "Que​​ faire de l’IA dans​​​‌ le champ de la​ création?", Journée d'étude ECCLA-Esadse,​‌ 28/01/2025.
• Rémi Eyraud :​​ 30 minutes of conference​​​‌ followed by 3h of​ workshop on "Centres sociaux​‌ et IA", centres sociaux​​ Loire et Haute-Loire, 70​​​‌ persons, 8/04/2025.
• Rémi Eyraud​ : 2h of workshop​‌ titled "Explorons l'IA", Fête​​ de la science, Explora​​​‌ Saint-Etienne, collégiens de 4ème​ et 3ème, 30 teens,​‌ 10/10/2025.

International journals

• 7 article​​​‌G.Guillem Bonafos,‌ F.Flavie Vial,‌​‌ C.Clément Cornec,​​ L.Léo Papet,​​​‌ L.Lény Lego,‌ J.Jessica Audouard,‌​‌ A.Aurélie de Vito​​, D.David Reby​​​‌, J.Jérémy Rouch‌, R.Rémi Emonet‌​‌, H.Hugues Patural​​ and N.Nicolas Mathevon​​​‌. Premature baby cries‌ remain different from those‌​‌ of full-term babies.​​Journal of the Acoustical​​​‌ Society of America157‌62025, 4175‌​‌ - 4183HALDOI​​
• 8 articleD.Dorian​​​‌ Llavata, E.Eleonora‌ Cagli, R.Rémi‌​‌ Eyraud, V.Vincent​​ Grosso and L.Lilian​​​‌ Bossuet. Unsupervised horizontal‌ attacks against public-key primitives‌​‌ with DCCA.IACR​​ Communications in Cryptology2​​​‌1April 2025HAL‌DOI
• 9 articleR.‌​‌Robin Mermillod-Blondin, N.​​Nicolas Dalloz, A.​​​‌Alain Tremeau, A.‌ W.Aldi Wista Fadhilah‌​‌, R.Rémi Emonet​​ and N.Nathalie Destouches​​​‌. Image quality metrics‌ for restricted gamut images‌​‌ produced by laser-induced printing​​​‌ on plasmonic films.​ Journal of the European​‌ Optical Society : Rapid​​ publications2122025​​​‌, 36HALDOI​back to text
• 10​‌ articleL.Lana Minier​​, J.Jérémy Rouch​​​‌, B.Bamdad Sabbagh​, F.Frédéric Bertucci​‌, E.Eric Parmentier​​, D.David Lecchini​​​‌, F.Frédéric Sèbe​, N.Nicolas Mathevon​‌ and R.Rémi Emonet​​. Visualization and quantification​​​‌ of coral reef soundscapes​ using CoralSoundExplorer software.​‌PLoS Computational Biology21​​4April 2025,​​​‌ e1012050HALDOI
• 11​ articleS.Sara Venturini​‌, M.Marianna de​​ Santis, J.Jordan​​​‌ Patracone, F.Francesco​ Rinaldi, S.Saverio​‌ Salzo and M.Martin​​ Schmidt. Relax and​​​‌ penalize: a new bilevel​ approach to mixed-binary hyperparameter​‌ optimization.Transactions on​​ Machine Learning Research Journal​​​‌February 2025HALback​ to text

International peer-reviewed​‌ conferences

• 12 inproceedingsQ.​​Quentin Bertrand, J.​​​‌ A.Juan Agustin Duque​, E.Emilio Calvano​‌ and G.Gauthier Gidel​​. Self-Play Q-Learners Can​​​‌ Provably Collude in the​ Iterated Prisoner's Dilemma.​‌Proceedings of the 42nd​​ International Conference on MachineLearning,​​​‌ Vancouver, Canada. PMLR 267,​ 2025International Conference on​‌ Machine LearningVancouver (BC),​​ CanadaJuly 2025HAL​​​‌back to text
• 13​ inproceedingsQ.Quentin Bertrand​‌, A.Anne Gagneux​​, M.Mathurin Massias​​​‌ and R.Rémi Emonet​. On the Closed-Form​‌ of Flow Matching: Generalization​​ Does Not Arise from​​​‌ Target Stochasticity.NeurIPS​ 2025NeurIPS 2025 -​‌ 39th Annual Conference on​​ Neural Information Processing Systems​​​‌San Diego (CA), United​ StatesDecember 2025HAL​‌
• 14 inproceedingsA.Antoine​​ Caradot, R.Rémi​​​‌ Emonet, A.-R.Abdel-Rahim​ Mezidi, A.Amaury​‌ Habrard and M.Marc​​ Sebban. Méthode de​​​‌ quadrature pour les PINNs​ fondée théoriquement sur la​‌ hessienne des résiduels.​​30ème Colloque sur le​​​‌ traitement du signal et​ des images GRETSIStrasbourg,​‌ FranceAugust 2025HAL​​
• 15 inproceedingsS.Stéphane​​​‌ Chrétien, B.Ben​ Gao, J.Jordan​‌ Patracone and O.Olivier​​ Alata. Using Signatures​​​‌ and Koopman Operator to​ Learn Nonlinear Dynamics.​‌7th International Conference of​​ Geometric Science of Information​​​‌ - GSI 2025GSI'25​ - 7th International Conference​‌ on Geometric Science of​​ Informationsaint-malo, France2025​​​‌HAL
• 16 inproceedingsB.​Ben Gao, J.​‌Jordan Patracone, S.​​Stephane Chrétien and O.​​​‌Olivier Alata. Conformal​ Online Learning of Deep​‌ Koopman Linear Embeddings.​​In Advances in Neural​​​‌ Information Processing SystemsNeurIPS​ 2025 - 39th Annual​‌ Conference on Neural Information​​ Processing SystemsSan Diego​​​‌ (California), United States2025​HAL
• 17 inproceedingsB.​‌Ben Gao, J.​​Jordan Patracone, S.​​​‌Stéphane Chrétien and O.​Olivier Alata. When​‌ to Learn: Conformal Scores​​ as Online Update Criteria​​​‌.Workshop on Predictions​ and Uncertainty at COLT​‌ 2025Lyon, FranceJune​​ 2025HAL
• 18 inproceedings​​​‌M.Mickaël Gault,​ P.Pierre Bachaud,​‌ B.Benoît Celse,​​ R.Rémi Emonet and​​​‌ M.Marc Sebban.​ Physics-Informed Machine Learning for​‌ Modeling CO2 Capture from​​ Scarce Data.37th​​ IEEE International Conference on​​​‌ Tools with Artificial Intelligence‌2025 IEEE 37th International‌​‌ Conference on Tools with​​ Artificial Intelligence (ICTAI)Athènes,​​​‌ GreeceIEEENovember 2025‌, 1436-1442HALDOI‌​‌back to text
• 19​​ inproceedingsA.-R.Abdel-Rahim Mezidi​​​‌, J.Jordan Patracone‌ and A.Amaury Habrard‌​‌. Douglas-Rachford Splitting for​​ Hybrid Differentiable Models.​​​‌Workshop on Differentiable Systems‌ and Scientific Machine Learning‌​‌ (DiffSys)Copenhagen, DenmarkDecember​​ 2025HAL
• 20 inproceedings​​​‌A.-R.Abdel-Rahim Mezidi,‌ J.Jordan Patracone,‌​‌ S.Saverio Salzo,​​ A.Amaury Habrard,​​​‌ M.Massimiliano Pontil,‌ R.Rémi Emonet and‌​‌ M.Marc Sebban.​​ A Bregman Proximal Viewpoint​​​‌ on Neural Operators.‌International Conference on Machine‌​‌ LearningVancouver, Canada2025​​HAL

Conferences without proceedings​​​‌

• 21 inproceedingsF.Fayad‌ Ali Banna, E.‌​‌Eduardo Brandão, A.​​Anthony Nakhoul, R.​​​‌Rémi Emonet, M.‌Marc Sebban and J.-P.‌​‌Jean-Philippe Colombier. Learning​​ dynamics in ultrafast laser​​​‌ self-organization.Progress in‌ Ultrafast Laser Modifications of‌​‌ MaterialsVillars-sur-Ollon, FranceJune​​ 2025HALback to​​​‌ text
• 22 inproceedingsG.‌Guillem Bonafos, J.‌​‌Jérémy Rouch, L.​​Lény Lego, D.​​​‌David Reby, H.‌Hugues Patural, N.‌​‌Nicolas Mathevon and R.​​Rémi Emonet. Speech​​​‌ transformer models for extracting‌ information from baby cries‌​‌.Workshop on Child​​ Computer Interaction - WOCCI​​​‌ 2025Nijmegen, NetherlandsISCA‌August 2025, 11‌​‌ - 15HALDOI​​
• 23 inproceedingsA.Antoine​​​‌ Caradot, R.Rémi‌ Emonet, A.Amaury‌​‌ Habrard, A.-R.Abdel-Rahim​​ Mezidi and M.Marc​​​‌ Sebban. Provably Accurate‌ Adaptive Sampling for Collocation‌​‌ Points in Physics-informed Neural​​ Networks.European Conference​​​‌ on Machine Learning and‌ Principles and Practice of‌​‌ Knowledge Discovery in Databases​​16017Lecture Notes in​​​‌ Computer SciencePorto, Portugal‌Springer Nature SwitzerlandSeptember‌​‌ 2025, 19-37HAL​​DOI
• 24 inproceedingsB.​​​‌Ben Gao, J.‌Jordan Patracone, O.‌​‌Olivier Alata and S.​​Stéphane Chrétien. Apprentissage​​​‌ incrémental de l'opérateur de‌ Koopman pour systèmes non-autonomes‌​‌ via prédiction conforme.​​Colloque GRETSI’25Strabourg, France​​​‌August 2025HAL
• 25‌ inproceedingsT.Thibault Girardin‌​‌, N.Nathalie Destouches​​ and A.Amaury Habrard​​​‌. Contextual Hypernetwork for‌ Adaptive Prediction of Laser-Induced‌​‌ Colors on Quasi-Random Plasmonic​​ Metasurfaces.European Conference​​​‌ on Machine Learning and‌ Principles and Practice of‌​‌ Knowledge Discovery in Databases​​Porto (Portugal), FranceSeptember​​​‌ 2025HALback to‌ text

Reports & preprints‌​‌

• 26 miscF.Fayad​​ Ali Banna, E.​​​‌Eduardo Brandão, A.‌Anthony Nakhoul, R.‌​‌Rémi Emonet, M.​​Marc Sebban and J.-P.​​​‌Jean-Philippe Colombier. Photonic‌ Self-Learning in Ultrafast Laser-Induced‌​‌ Complexity.March 2025​​HALback to text​​​‌
• 27 miscF.Fayad‌ Ali Banna, A.‌​‌Antoine Caradot, E.​​Eduardo Brandão, J.-P.​​​‌Jean-Philippe Colombier, R.‌Rémi Emonet and M.‌​‌Marc Sebban. Unrolled-SINDy:​​ A Stable Explicit Method​​​‌ for Non linear PDE‌ Discovery from Sparsely Sampled‌​‌ Data.October 2025​​HAL
• 28 miscH.​​​‌Hind Atbir, F.‌Farah Cherfaoui, G.‌​‌Guillaume Metzler, E.​​​‌Emilie Morvant and P.​Paul Viallard. PAC-Bayesian​‌ Bounds on Constrained f-Entropic​​ Risk Measures.October​​​‌ 2025HAL
• 29 misc​A.Antoine Caradot and​‌ Z.Zongzhu Lin.​​ Differential graded algebras with​​​‌ divided powers and homotopy​ Lie algebras.November​‌ 2025HAL
• 30 misc​​S.Stéphane Chrétien,​​​‌ B.Ben Gao,​ J.Jordan Patracone,​‌ A.-R.Abdel-Rahim Mezidi and​​ O.Olivier Alata.​​​‌ Tukey-Median of Means-Gradients for​ Langevin Dynamics: a Robust​‌ Approach to Fitting Machine​​ Learning Models.October​​​‌ 2025HAL
• 31 misc​T.Tiantian Feng,​‌ B. M.Brandon M​​ Booth, K.Karel​​​‌ Mundnich, E.Emily​ Zhou, B.Benjamin​‌ Girault, K.Kristina​​ Lerman and S.Shrikanth​​​‌ Narayanan. TILES-2018 Sleep​ Benchmark Dataset: A Longitudinal​‌ Wearable Sleep Data Set​​ of Hospital Workers for​​​‌ Modeling and Understanding Sleep​ Behaviors.July 2025​‌HAL

Other scientific publications​​

• 32 miscA.Anne​​​‌ Gagneux, S.Ségolène​ Martin, R.Rémi​‌ Emonet, Q.Quentin​​ Bertrand and M.Mathurin​​​‌ Massias. A Visual​ Dive into Conditional Flow​‌ Matching.April 2025​​HAL