OPIS

OPIS - 2025

2025Activity‌ reportProject-TeamOPIS

RNSR: 201923238F

Research center Inria‌ Saclay Centre at Université Paris-Saclay
In partnership with:‌Université Paris-Saclay
Team name: OPtImization for large Scale‌ biomedical data

Creation of the Project-Team: 2019 May‌ 01

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

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

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

Keywords

Computer Science and Digital Science‌

A3.4. Machine learning and‌ statistics
A6.2. Scientific computing,‌‌ Numerical Analysis & Optimization
A6.2.4. Statistical methods
A6.2.6.‌ Optimization
A8.2. Optimization
A8.7.‌ Graph theory
A9.2. Machine‌‌ learning
A9.2.1. Supervised learning
A9.2.2. Unsupervised learning
A9.2.3.‌ Reinforcement learning
A9.2.4. Optimization‌ and learning
A9.2.5. Bayesian‌‌ methods
A9.2.6. Neural networks
A9.2.8. Deep learning
A9.3.‌ Signal processing
A9.7. AI‌ algorithmics

1 Team members, visitors, external collaborators‌

Research Scientist

Emilie Chouzenoux‌ [Team leader,‌‌ INRIA, Senior Researcher, HDR]

Faculty‌ Members

Marc Castella [‌TELECOM SUD PARIS,‌‌ Professor]
Mounir Kaaniche [SORBONNE PARIS NORD‌, Professor]
Fragkiskos‌ Malliaros [CENTRALESUPELEC,‌‌ Professor, HDR]
Nora Ouzir [CENTRALESUPELEC‌, Associate Professor]‌
Jean-Christophe Pesquet [CENTRALESUPELEC‌‌, Professor, HDR]
Antonio Silveti-Falls [‌CENTRALESUPELEC, Associate Professor‌]
Hugues Talbot [‌‌CENTRALESUPELEC, Professor, HDR]
Maria Vakalopoulou‌ [CENTRALESUPELEC, Associate‌ Professor]

Post-Doctoral Fellows‌‌

Ishak Ayad [CENTRALESUPELEC, Post-Doctoral Fellow,‌ from Nov 2025]‌
Daniele Malitesta [CENTRALESUPELEC‌‌, Post-Doctoral Fellow, from Mar 2025]‌

PhD Students

Alix Chazottes‌ [CENTRALESUPELEC]
Clement‌‌ Cosserat [INRIA]
Paul Delage [INRIA‌, CIFRE, from‌ Oct 2025, SOCOTEC‌‌]
Eve Delegue [CENTRALESUPELEC]
Hafsa El‌ Herichi [CENTRALESUPELEC,‌ CIFRE, SAFRAN]‌‌
Yelizaveta Falkouskaya [CENTRALESUPELEC, from Oct 2025‌]
Adam Ghalem [‌CENTRALESUPELEC, CIFRE,‌‌ NOKIA]
Thomas Guilmeau [CENTRALESUPELEC, until‌ Jan 2025]
Vuk‌ Ignjatovic [CENTRALESUPELEC,‌‌ CIFRE, GE HEALTHCARE]
Loïc Le Bescond‌ [CENTRALESUPELEC]
Daniele‌ Malitesta [CENTRALESUPELEC,‌‌ until Feb 2025]
Shuai Mao [CENTRALESUPELEC‌]
Vahan Martirosyan [‌CENTRALESUPELEC]
Imed Moussa‌‌ [INRIA, CIFRE, SAFRAN]
Ana‌ Neacsu [CENTRALESUPELEC]‌
Marian Negru [CENTRALESUPELEC‌‌]
Eline Pot [CENTRALESUPELEC, CIFRE,‌ FRAMATOME]
Ali Ramlaoui‌ [CENTRALESUPELEC, CIFRE‌‌, from Jun 2025, ENTALPIC]
Nicolas‌ Salvy [INRIA]‌
Aymen Sardroui [CENTRALESUPELEC‌‌]
Ludovic Trautmann [INRIA, from Sep‌ 2025]
Antoine Vialle‌ [IP PARIS,‌‌ from Nov 2025]

Technical Staff

Samy Ferrat‌ [INRIA, Engineer‌, from Oct 2025‌‌]
Tristan Portugues [INRIA, Engineer,‌ from Feb 2025 until‌ Nov 2025]
Ludovic‌‌ Trautmann [INRIA, Engineer, until Aug‌ 2025]
Quentin Vanderbecq‌ [INRIA, from‌‌ Nov 2025, Poste d'Accueil APHP]

Interns‌ and Apprentices

Maxence Adly‌ [INRIA, Intern‌‌, until Feb 2025‌]
Bruno Amorim De Araujo [INRIA,‌ Intern, from Apr 2025 until Sep 2025‌]
Mohamed Salim Ben Omrane [CENTRALESUPELEC,‌ Intern, from Apr 2025 until Sep 2025‌]
Idriss Benkirane [CENTRALESUPELEC, until Feb‌ 2025]
Alexandre Bertot [CENTRALESUPELEC, from‌ Aug 2025]
Benjamin Clene [CENTRALESUPELEC,‌ until Jan 2025]
Enzo Dehedim [CENTRALESUPELEC‌, from Jun 2025 until Jul 2025]‌
Eve Delegue [CENTRALESUPELEC, from Mar 2025‌ until Aug 2025]
Yassine Elammari [CENTRALESUPELEC‌, from Jul 2025]
Luis Evrard [‌CENTRALESUPELEC, from Apr 2025 until Nov 2025‌]
Ismail Hatim [CENTRALESUPELEC, until Oct‌ 2025]
Mohammad Mehdi Kalla [INRIA,‌ Apprentice]
Matthieu Merigot–Lombard [INRIA, Intern‌, from May 2025 until Oct 2025]‌
Aravind Subramanian [CENTRALESUPELEC, from May 2025‌ until Jun 2025]
Ziu Wu [CENTRALESUPELEC‌, from May 2025 until Oct 2025]‌
Fatma Zahra [CENTRALESUPELEC, from Apr 2025‌ until Aug 2025]
Bilal Zidna [INRIA‌, Intern, from Mar 2025 until Aug‌ 2025]

Administrative Assistants

Jana Dutrey [CENTRALESUPELEC‌]
Bamissa Sangare [INRIA, from Sep‌ 2025]
Joyce Soares Brito [INRIA]‌

Visiting Scientists

Ilias Kounis [APHP, from‌ Nov 2025]
Subhajit Saha [TCG CREST‌, from Feb 2025 until Apr 2025]‌
Vlad Vasilescu [UNIVERSITY POLITEHNICA OF BUCHAREST,‌ from Feb 2025]

2 Overall objectives

Mathematical‌ optimization is the key to solving many problems‌ in science, based on the observation that physical‌ systems obey a general principle of least action.‌ While some problems can be solved analytically, many‌ more can only be solved via numerical algorithms.‌ Research in this domain has been steadily ongoing‌ for decades.

In addition, many fields such as‌ medecine continue to benefit from considerable improvements in‌ data acquisition technology, based on sophisticated tools from‌ optics and physics (e.g., new laser sources in‌ microscopy, multi-coil systems in MRI, novel X-ray schemes‌ in mammography, etc). This evolution is expected to‌ yield significant improvements in terms of data resolution,‌ making the interpretation and analysis of the results‌ easier and more accurate for the practitioner. The‌ large amounts of generated data must be analyzed‌ by sophisticated optimization tools so that, in recent‌ years, optimization has become a main driving force‌ fostering significant advances in data processing. Previously hidden‌ or hard to extract information can be pried‌ from massive datasets by modern recovery and data‌ mining methods. At the same time, automated decision‌ and computer-aided diagnoses are made possible through optimal‌ learning approaches.

However, major bottlenecks still exist. Recent‌ advances in instrumentation techniques come with the need‌ to minimize functions involving an increasingly large number‌ of variables (at least one billion variables in‌ 3D digital tomography modality), and with increasingly complex‌ mathematical structure. The computational load for solving these problems may be too‌ high for even state-of-the-art‌ algorithms. New algorithms must‌‌ be designed with computational scalability, robustness, and versatility‌ in mind. In particular,‌ the following severe requirements‌‌ must be fulfilled: (i) ability to tackle high-dimensional‌ problems in a reasonable‌ computation time; (ii) low-requirements‌‌ in terms of memory usage; (iii) robustness to‌ incomplete or unreliable information;‌ (iv) adaptivity to statistically‌‌ varying environments; (v) resilience to latency issues arising‌ in architectures involving multiple‌ computing units.

These difficulties‌‌ are compounded in the medical and biomedical areas.‌ In these contexts, datasets‌ are not easily available‌‌ due to patient confidentiality and/or instrument limitations. Moreover,‌ high-level expertise is necessary‌ to interpret the data‌‌ which can be of very high dimension. Finally,‌ the developed analysis methods‌ must be reliable and‌‌ interpretable by the medical/biomedical community.

The objective of‌ the OPIS project is‌ to design advanced optimization‌‌ methods for the analysis and processing of large‌ and complex data. Applications‌ to inverse problems and‌‌ machine learning tasks in biomedical imaging are major‌ outcomes of this research‌ project. We seek optimization‌‌ methods able to tackle data with both a‌ large sample-size (“big $N‌$ " e.g., $N =‌‌ 10^{9}$ ) and/or many measurements (“big $P‌$ " e.g., $P =‌ 10^{4}$ ). The‌‌ methodologies to be explored are grounded on nonsmooth‌ functional analysis, fixed point‌ theory, parallel/distributed strategies, and‌‌ neural networks. The new optimization tools that are‌ developed are set in‌ the general framework of‌‌ graph signal processing, encompassing both regular graphs (e.g.,‌ images) and non-regular graphs‌ (e.g., gene regulatory networks).‌‌

More specifically, three main research avenues are explored,‌ namely:

Accelerated algorithms for‌ solving high-dimensional continuous optimization‌‌ problems ;
Optimization over graphs ;
Toward more‌ understandable deep learning.

In‌ summary, the specificity of‌‌ OPIS is to address problems involving high-dimensional biomedical‌ data, e.g. 3D CT,‌ PET, ultrasound images, and‌‌ MRI, by making use of advanced computational optimization‌ methods.

3 Research program‌

3.1 Accelerated algorithms for‌‌ solving high-dimensional continuous optimization problems

Variational problems requiring‌ the estimation of a‌ huge number of variables‌‌ have now to be tackled, especially in the‌ field of 3D reconstruction/restoration‌ (e.g. $\geq 10^{9‌‌}$ variables in 3D imaging). In addition to the‌ curse of dimensionality, another‌ difficulty to overcome is‌‌ that the cost function usually reads as the‌ sum of several loss/regularization‌ terms, possibly composed with‌‌ large-size linear operators. These terms can be nonsmooth‌ and/or nonconvex, as they‌ may serve to promote‌‌ the sparsity of the sought solution in some‌ suitable representation (e.g. a‌ frame) or to fufill‌‌ some physical constraints. In such a challenging context,‌ there is a strong‌ need for developing fast‌‌ parallelized optimization agorithms for which sound theoretical guarantees‌ of convergence can be‌ established. We explore deterministic‌‌ and stochastic approaches based on proximal tools, MM‌ (Majorization-Minimization) strategies, and trust‌ region methods. More generally,‌‌ we are interested in using fixed point methods‌ which provide a simplifying‌ and unifying framework to‌‌ model, analyze, and solve‌ a great variety of problems. They constitute a‌ natural environment to explain the behavior of advanced‌ convex optimization methods as well as of recent‌ nonlinear methods in data science which are formulated‌ in terms of paradigms that go beyond minimization‌ concepts and involve constructs such as Nash equilibria‌ or monotone inclusions. Because of the versatility of‌ the methods, a wide range of applications in‌ image recovery are considered, such as dynamic positron‌ emission tomography (PET), 3D ultrasound imaging, and two-photon‌ microscopy. For example, in dynamic PET imaging (collaboration‌ with CEA - Biomaps), we must solve a‌ tomographic ill-posed inverse problem with Poisson noise of‌ particularly high intensity. Our goal is to devise‌ an efficient dose-dependent regularization scheme adapted to the‌ dynamic protocol, with interpretable learned hyperparameters. In two-photon‌ microscopy (collaboration with XLIM), our objective is to‌ provide effective numerical solutions to improve the 3D‌ resolution of the microscope, especially when cheap laser‌ sources are used, with applications to bacteria imaging,‌ and muscle disease screening.

3.2 Optimization over graphs‌

Graphs and hypergraphs are rich data structures for‌ capturing complex, possibly irregular, dependencies in multidimensional data.‌ Coupled with Markov models, they constitute the backbones‌ of many techniques used in computer vision. Optimization‌ is omnipresent in graph processing. Firstly, it allows‌ the structure of the underlying graph to be‌ inferred from the observed data, when the former‌ is hidden. Second, it permits to develop graphical‌ models based on the prior definition of a‌ meaningful cost function. This leads to powerful nonlinear‌ estimates of variables corresponding to unknown weights on‌ the vertices and/or the edges of the graph.‌ Tasks such as partitioning the graph into subgraphs‌ corresponding to different clusters (e.g., communities in social‌ networks) or graph matching, can effectively be performed‌ within this framework. Finally, graphs by themselves offer‌ flexible structures for formulating and solving optimization problems‌ in an efficient distributed manner. On all these‌ topics, our group has acquired a long-term expertise‌ that we plan to further strengthen. In terms‌ of applications, novel graph mining methods are proposed‌ for gene regulatory and brain network analysis. For‌ example, we plan to develop sophisticated methods for‌ better understanding the gene regulatory network of various‌ microscopic fungi, in order to improve the efficiency‌ of the production of bio-fuels (collaboration with IFP‌ Energies Nouvelles).

3.3 Toward more understandable deep learning‌

Nowadays, deep learning techniques efficiently solve supervised tasks‌ in classification or regression by utilizing large amounts‌ of labeled data and the powerful high level‌ features that they learn by using the input‌ data. Their good performance has caught the attention‌ of the optimization community since currently these methods‌ offer virtually no guarantee of convergence, stability or‌ generalization. Deep neural networks are optimized through a‌ computationally intensive engineering process via methods based on‌ stochastic gradient descent. These methods are slow and‌ they may not lead to relevant local minima.‌ Thus, more efforts must be dedicated in order to improve the training‌ of deep neural networks‌ by proposing better optimization‌‌ algorithms applicable to large-scale datasets. Beyond optimization, incorporating‌ some structure in deep‌ neural networks permits more‌‌ advanced regularization than the current methods. This should‌ reduce their complexity, as‌ well as allow us‌‌ to derive some bounds regarding generalization. For example,‌ many signal processing models‌ (e.g. those based on‌‌ multiscale decompositions) exhibit some strong correspondence with deep‌ learning architectures, yet they‌ do not require as‌‌ many parameters. One can thus think of introducing‌ some supervision into these‌ models in order to‌‌ improve their performance on standard benchmarks. A better‌ mathematical understanding of these‌ methods permits to improve‌‌ them, but also to propose some new models‌ and representations for high-dimensional‌ data. This is particularly‌‌ interesting in settings such as the diagnosis or‌ prevention of diseases from‌ medical images, because they‌‌ correspond to critical applications where the made decision‌ is crucial and needs‌ to be interpretable. One‌‌ of the main applications of this work is‌ to propose robust models‌ for the prediction of‌‌ the outcome of cancer immunotherapy treatments from multiple‌ and complementary sources of‌ information: images, gene expression‌‌ data, patient profile, etc (collaboration with Institut Gustave‌ Roussy).

4 Application domains‌

4.1 Robustness of AI‌‌

4.1.1 Robust few-shot learning

Participants:Nora Ouzir and‌ Jean-Christophe Pesquet

We aim‌ to advance the frontiers‌‌ of few-shot learning by integrating the adaptability of‌ few-shot frameworks with the‌ rigour of classical robust‌‌ statistical methods. By systematically investigating how quality, volume,‌ and diversity of training‌ data shape model performance,‌‌ the research seeks to overcome a fundamental bottleneck‌ in AI-driven healthcare: the‌ scarcity of high-quality annotated‌‌ data. Beyond advancing methodological understanding, the research aspires‌ to transform clinical practice‌ by enabling accurate and‌‌ reliable AI diagnostics with just minimal labelled data,‌ informing data-efficient strategies for‌ large-scale medical imaging, and‌‌ ultimately accelerating the adoption of AI technologies to‌ improve patient outcomes across‌ diverse healthcare systems.

4.1.2‌‌ Design of robust neural networks in safety critical‌ industrial domains

Participants:‌ Jean-Christophe Pesquet , Emilie‌‌ Chouzenoux

One of the main challenges faced today‌ by companies like Thales‌ or Schneider Electric designing‌‌ advanced industrial systems, is to ensure the safety‌ of new generations of‌ products based on the‌‌ use of neural networks. Since 2013, neural networks‌ have been shown to‌ be sensitive to adversarial‌‌ perturbations. Deep neural networks can thus be fooled,‌ in an intentional (security‌ issue) or in undeliberate‌‌ manner (safety issue), which raises a major robustness‌ concern for safety-critical systems‌ which need to be‌‌ certified by an independent certification authority prior to‌ any entry into production/operation.‌ Tech- niques based on‌‌ mathematical proofs of robustness are generally preferred by‌ industrial safety experts since‌ they enable a safe-by-design‌‌ approach that is more efficient than a robustness‌ verification activity done a‌ posteriori with a necessarily‌‌ bounded effort. Among the possible mathematical approaches, we‌ focus on those relying‌ upon the analysis of‌‌ the Lipschitz properties of‌ neural networks 9. Such properties play a‌ fundamental role in the understanding of the internal‌ mechanisms governing these complex nonlinear systems. Besides, they‌ make few assumptions on the type of non-linearities‌ used and are thus valid for a wide‌ range of networks.

4.1.3 Certification of segmentation networks‌

Participants: Othmane Laousy , Maria Vakalopoulou (Collaboration:‌ G. Chassagnon and M.-P. Revel, Paris Cité University‌ ; N. Paragios, Therapanacea ; A. Araujo, S.‌ Garg and F. Khorram, New York University)

The‌ robustness of image segmentation has been an important‌ research topic in the past few years as‌ segmentation models have reached production-level accuracy. However, like‌ classification models, segmentation models can be vulnerable to‌ adversarial perturbations, which hinders their use in critical-decision‌ systems like healthcare or autonomous driving. Recently, randomized‌ smoothing has been proposed to certify segmentation predictions‌ by adding Gaussian noise to the input to‌ obtain theoretical guarantees. However, this method exhibits a‌ trade-off between the amount of added noise and‌ the level of certification achieved. In this topic,‌ we address the problem of certifying segmentation prediction‌ using a combination of randomized smoothing and diffusion‌ models. We challenge our methods in both general‌ computer vision and medical imaging dataset.

4.2 Radiology,‌ hyper-progressive disease and immunotherapy

4.2.1 Realtime flow estimation‌ for interventional doppler ultrafast ultrasound

Participants: Nora Ouzir‌ , Jean-Christophe Pesquet (Collaboration: IRIT laboratory)

Assessing blood‌ flow and vascular structures is crucial for treating‌ various conditions, including brain tumours and cardiovascular disorders.‌ During surgical procedures, the successful removal of a‌ tumour relies on accurately defining the boundary between‌ the tumour and the surrounding vital brain tissues.‌ Traditional methods for separating blood flow from tissues‌ typically employ filtering techniques, often utilizing SVD. More‌ recent approaches, like DRPCA, tackle an inverse problem‌ that incorporates blood sparsity along with the low-rank‌ structure of tissues. In our recent work 24‌, we expanded this method to account for‌ tissue motion, developing an algorithm that jointly estimates‌ blood flow, tissues, and their movements. This advancement‌ results in more accurate blood flow estimations and‌ significantly reduces sensitivity to motion. Currently, we are‌ considering several challenges hindering the clinical applicability of‌ 24, including slow execution.

4.2.2 Imaging radiomics‌ and pathomics to assess response to treatment

Participants:‌

Maria Vakalopoulou , Hugues Talbot (Collaboration: E. Deutsh,‌ Institut Gustave Roussy ; N. Paragios, Therapanacea)

The‌ response of patients with cancer to immunotherapy can‌ vary considerably, innovative predictors of response to treatment‌ are needed to improve treatment outcomes. We aimed‌ to develop and independently validate radiomics-based biomarkers of‌ tumour-infiltrating cells in patients included in trials of‌ the two most common, recent immunotherapy treatments: anti-programmed‌ cell death protein (PD)-1 or anti-programmed cell death‌ ligand 1 (PD-L1) monotherapy. We also aimed to‌ evaluate the association between the biomarker, and tumour‌ immune phenotype and clinical outcomes of these patients.‌

However, sometimes, not only do patient respond poorly,‌ but immunotherapy seems to make things worse. Some patients see they tumoral‌ load increase significantly faster‌ after immunotherapy is started.‌‌ These patients are called “hyper-progressors”. One of our‌ project has been to‌ clearly define and detect‌‌ this class of patients. This is so novel‌ that the very notion‌ of hyperprogressive patient was‌‌ still controversial when our work was published, but‌ is accepted now.

In‌ this axis we investigate‌‌ powerful representations for radiological and pathological data that‌ could be associated with‌ interesting and important clinical‌‌ questions.

4.2.3 Analysis of histopathology images for cancer‌ treatment

Participants: Ségolène‌ Martin , Nora Ouzir‌‌ , Jean-Christophe Pesquet , Aymen Sardroui (Collaboration: A.‌ Laurent-Bellue, C. Guettier: APHP,‌ Hôpital du Kremlin-Bicêtre ;‌‌ A. Beaufrère, K. Mondet, V. Paradis, APHP, Hôpital‌ Beaujon)

The core focus‌ of our research revolves‌‌ around scrutinizing cancer through the utilization of digital‌ slide images resulting from‌ biopsies or surgical resection.‌‌ Our exploration stands at the intersection of cutting-edge‌ AI technology and its‌ invaluable potential for advancing‌‌ precision medicine, and more particularly liver cancer (hepatocellular‌ carcinoma and intrahepatic cholangiocarcinoma)‌ diagnosis and treatment. The‌‌ challenges to be solved are related to the‌ limited number of available‌ annotated data and the‌‌ large-size of whole slide images (WSIs).

4.2.4 Vision,‌ machine learning and precision‌ medicine

Participants:Younes Belkouchi‌‌ , Loïc Le Bescond , Hugues Talbot (Collaboration:‌ N. Lassau, Institut Gustave‌ Roussy)

In March 2020,‌‌ the PRISM institute of Gustave-Roussy was launched. The‌ aim of this project,‌ funded for 5 years,‌‌ is to develop targeted treatments that are more‌ likely to work on‌ specific patients.

The mission‌‌ of this “second-generation” precision medicine centre will be‌ to model cancer on‌ an individual scale by‌‌ creating numerical avatars of tumours. The aim is‌ to identify patients with‌ the most aggressive cancers‌‌ very early in the disease, without waiting for‌ relapses, in order to‌ offer them the most‌‌ appropriate treatment from the start of treatment, using‌ the huge volume of‌ clinical, biological and molecular‌‌ data and their analysis by artificial intelligence. PRISM‌ will conduct large-scale clinical‌ studies and develop molecular‌‌ analysis technologies and data analysis methods.

Coordinated by‌ Professor Fabrice André, Research‌ Director of Gustave Roussy,‌‌ Inserm Research Director and Professor at Paris-Saclay University,‌ Prism aims to revolutionize‌ the understanding of the‌‌ molecular and biological mechanisms of cancer development and‌ progression through artificial intelligence.‌ Based on increasingly rich‌‌ data of various types (clinical, genomic, microbiological, imaging,‌ etc.), learning algorithms make‌ it possible to develop‌‌ finer diagnostic and prognostic tools, and thus to‌ propose therapies that are‌ personalised according to the‌‌ characteristics of the individual.

Funded by the French‌ National Research Agency, PRISM‌ received the IHU label‌‌ in 2018, followed by the National Center for‌ Precision Medicine label.

4.2.5‌ Physics-informed, generative models for‌‌ heart left ventricle perfusion analysis

Participants: Raoul‌ Salle de Chou, Hugues‌ Talbot (Collaboration: I. Vignon-Clementel,‌‌ SIMBIOTX Team Inria ; L. Najman, Université Gustave-Eiffel,‌ L. Papamanolis Stanford university,‌ USA, California)

Coronary arteries‌‌ feed the heart muscles‌ with nutrients and oxygen. As such, they are‌ some of the most critical blood vessel in‌ the entire body. Coronary disease is difficult to‌ diagnose especially when it affects the smaller branches‌ of these vessels, because direct imaging of these‌ vessels is infeasible with current medical imaging technology.‌ Instead, blood perfusion through the myocardium can be‌ imaged and is correlated with both arterial and‌ myocardium disease. However, perfusion imaging is challenging, invasive‌ and expensive because it relies on radioactive tracers.‌

A previous model was developed for myocardial perfusion‌ simulation for coronary artery disease in [link] to‌ replace the actual exam with a numerical twin‌ and conduct it via simulations. The model aims‌ at reproducing $[^{15} O] H_{2‌} O$ PET imaging exam using only CT scans‌ as input. The simulation is based on :‌

the detection, segmentation and simulation of blood flow‌ through the coronary vessels visible on the injected‌ CT scan;
a patient-specific method for generating small‌ 3D vessels consistent with the vessels detected. The‌ rules for the growth of these vessels are‌ based on physiology and simulated blood flows;
a‌ perfusion simulation model that considers the myocardium as‌ a porous medium.

For this a linear Darcy‌ model is used to simulate blood flow through‌ the porous medium. However, in addition to a‌ high computational cost, the simulation fails to accurately‌ reproduce some diseases, particularly those that affect medium-size‌ coronary branches.

The main goal of this project‌ is to combine Machine Learning (ML) methods with‌ physical simulations, in order to improve the current‌ simulation pipeline. ML algorithms are used to learn‌ from PET imaging exams while being guided by‌ simulation hypothesis, thereby diminishing the dependency on patient‌ data. To achieve this, each part of the‌ simulation is to be replaced by an ML‌ model. Following successful replication of simulation outcomes, the‌ model will undergoes refinement using patient data.

A‌ finite volume physics informed graph neural network was‌ developed to solve the Darcy equations on irregular‌ shapes serving as a substitute for the myocardium‌ component in perfusion simulation. Preliminary results indicate superior‌ performance of this model in terms of accuracy‌ and generalization compared to classical ML approaches. In‌ XX, we introduced a novel optimization framework for‌ the generation of the synthetic small vessels utilizing‌ the constructed constrained optimization (CCO) method. Our new‌ approach simulated similar 2D vascular trees as the‌ original CCO method in terms of morphometry while‌ producing better optimal solutions at lower computational cost.‌ This new approach is expected to be more‌ readily reproducible using ML methods compared to the‌ original CCO technique.

Additionally, work has been conducted‌ towards the determination of the myocardium perfusion regions.‌ Determining these regions, and their associated vessel is‌ a crucial step in current simulation pipeline. However,‌ the current calculation method is inaccurate and highly‌ sensitive to the resolution of segmented vessels. A‌ more robust and accurate model, employing graph neural networks (GNNs), has been‌ developed for the determination‌ of these regions.

4.3‌‌ Sparse inverse problems

4.3.1 Sparse signal processing in‌ chemistry

Participants:Emilie Chouzenoux‌ , Mohammad Mehdi Kalla‌‌ , Jean-Christophe Pesquet (Collaboration: L. Duval, IFPEN, Rueil‌ Malmaison)

Peak-signal retrieval is‌ a core challenge in‌‌ separative analytical chemistry (AC). For instance, in chromatography,‌ spectrometry, spectroscopy, peak localization,‌ amplitude, width or area‌‌ provide useful chemical quantitative information. We investigated this‌ problem through the deep‌ unrolling paradigm, in collaboration‌‌ with Dr. L. Duval, Research Engineer at IFP‌ Energies Nouvelles, France.

4.3.2‌ Image restoration for Multiplex-Coherent‌‌ Anti-Stokes Raman Scattering (M-CARS) spectroscopy

Participants:Samy Ferrat‌ , Ludovic Trautmann ,‌ Emilie Chouzenoux , Jean-Christophe‌‌ Pesquet (Collaboration: C. Lefort, XLIM, CNRS, Limoges ;‌ CHU Limoges)

Label-free discrimination‌ between protein families within‌‌ the same biological environment remains a significant challenge‌ for bioimaging. Through an‌ ongoing collaboration with physicists‌‌ from XLIM laboratory (CNRS, Limoges, France), and CHU‌ Limoges, funded by PIQ‌ project Khi-MalMa, we investigate‌‌ advanced mathematical and computational solutions for M-CARS image‌ restoration. In the preliminary‌ work 22, we‌‌ leverage the intrinsic differences in nonlinear optical responses‌ between proteins to propose‌ a new strategy that‌‌ allows qualitative label-free imaging and enhances the analytical‌ potential of M-CARS spectroscopy‌ for multifactorial biological studies.‌‌

4.3.3 Reconstruction approaches in PET imaging

Participants:Emilie‌ Chouzenoux , Alix Chazottes‌ , Ludovic Trautmann ,‌‌ Jean-Christophe Pesquet (Collaboration: F. Sureau, CEA BioMaps +‌ AAIMME project partners)

Positron‌ emission tomography (PET) is‌‌ a quantitative functional imaging modality used to track‌ the fate and/or dynamics‌ of a radiotracerpreviously injected‌‌ into a patient. This technique is particularly used‌ in oncology for diagnosis‌ and therapeutic monitoring, in‌‌ the study ofneurodegenerative diseases, and in pharmacology. In‌ dynamic PET, the temporal‌ evolution of the spatial‌‌ distribution of the radiotracerduring the examination is taken‌ into account for the‌ estimation of physiological parameters‌‌ allowing for a fine characterization of themolecular mechanisms‌ at play (receptor concentration,‌ absorption, dissociation constants, binding‌‌ potential, etc.). In the PhD thesis of Alix‌ Chazottes , in collaboration‌ between OPIS and CEA,‌‌ we aim to propose methodological developments in the‌ fields of optimization and‌ learning to address the‌‌ problem of robust dynamic PET reconstruction. In the‌ PhD thesis of Ludovic‌ Trautmann (ANR AAIMME), we‌‌ investigate the joint image reconstruction problem and estimation‌ of uncertainties,in the context‌ of PET imaging with‌‌ innovative time-of-flight detectors.

4.4 Graph mining applications

4.4.1‌ Geometric Graph Neural Networks‌ for molecular and chemical‌‌ systems

Participants:Fragkiskos Malliaros , Hugues Talbot (Collaboration:‌ A. Duval, Entalpic)

Graph‌ Neural Networks (GNNs) currently‌‌ constitute state-of-the-art models for solving prediction tasks on‌ graphs. Through the flexible‌ formulation of the message‌‌ passing mechanism, GNNs can learn informative latent representations‌ of graph entities at‌ different resolution levels (e.g.,‌‌ node-, edge-, graph-level). In many practical applications in‌ molecular and chemical systems,‌ the nodes of the‌‌ graph have associated geometric attributes (e.g., coordinates, velocities)‌ related to their position‌ in the 3D space.‌‌ In this context, geometric‌ graphs represent the interaction of atoms in the‌ 3D space, encapsulating a range of physical symmetries‌ such as rotations and translations. Existing GNN models‌ often overlook this aspect, rendering them ill-suited for‌ prediction tasks on geometric graphs. Recently, Geometric GNN‌ architectures tailored to respect physical symmetries have emerged‌ as flexible models of atomic systems. Through an‌ ongoing collaboration with Entalpic, we study geometric GNN‌ models, focusing both on design principles as well‌ as on practical applications in materials modeling (e.g.,‌ property prediction and molecule generation).

4.4.2 Graph machine‌ learning for spatiotemporal data

Participants:Fragkiskos Malliaros (Collaboration:‌ J. Castro-Correa, Mohsen Badiey, Univ. of Delaware ;‌ J. H. Giraldo, Télécom Paris)

Numerous real-world prediction‌ problems involve spatiotemporal data. For example, consider sensors‌ scattered across diverse geographical regions measuring environmental conditions‌ (e.g., temperature, pollution) or functional magnetic resonance imaging‌ (fMRI) data capturing brain activity. Both scenarios generate‌ data inherently rich in spatiotemporal structure, benefiting from‌ the relational inductive bias of graph-based modeling. In‌ an ongoing collaboration with the University of Delaware,‌ Télécom Paris, and La Rochelle Université, we have‌ introduced a methodology that leverages graph-based modeling, enabling‌ time series imputation with GNNs. Major challenges here‌ concern inducing temporal and relational smoothness assumptions into‌ the model as well as inferring the (often‌ unknown) graph structure. Furthermore, an intriguing aspect involves‌ enhancing spatiotemporal graph models with causal properties to‌ capture causal influence effects among entities.

4.4.3 Graph‌ representation learning for computational medicine

Participants:Emilie Chouzenoux‌ , Fragkiskos Malliaros (Collaboration: A. Majumdar, IIIT Delhi‌ ; B. Liu, D. Papadopoulos, G. Tsoumakas, A.‌ Papadopoulos, Aristotle Univ. of Thessaloniki )

Following the‌ Associate International Inria Team COMPASS (ended in 2023),‌ led by Emilie Chouzenoux and Dr. A. Majumdar‌ (IIIT Delhi), we pursue our research to investigate‌ new models and inference tools to understand and‌ predict optimal drug association, so as to tackle‌ real-life problems of computational drug discovery. We introduced‌ graph-based regularization techniques in order to incorporate expert‌ knowledge and metadata in matrix completion tasks arising‌ in the prediction of drug-virus and drug-drug associations.‌ Our recent works apply the proposed methodology to‌ drug repositioning 8.

The discovery of drug-target‌ interactions is also explored by Fragkiskos Malliaros ,‌ in collaboration with Aristotle University of Thessaloniki. Accurately‌ identifying reliable interactions among drugs and proteins via‌ computational methods, which typically leverage heterogeneous information retrieved‌ from diverse data sources, can boost the development‌ of effective pharmaceuticals. We have considered multi-layered network‌ modeling to handle diverse drug and target similarities,‌ introducing an optimization framework called Multiple similarity DeepWalk-based‌ Matrix Factorization (MDMF) for DTI prediction. Current efforts‌ in this direction involve leveraging Graph Neural Networks‌ and self-supervised learning tools.

4.4.4 Graph-based models for‌ multimodal patient representation learning

Participants:Fragkiskos Malliaros ,‌ Daniele Malitesta (Collaboration: N. Garcelon, A. Burgun, Institut‌ Imagine)

Fusing heterogeneous patient data – ranging from‌ demographic data to clinical measurements and genetic information‌ – is a critical step in precision medicine toward characterizing the progression‌ history of a disease‌ and learning predictive models‌‌ for diagnosis. In the context of the ANR‌ RHU project Innov4-ePiK (2024‌ – 2029) led by‌‌ the Imagine Institute of Genetic Diseases, we investigate‌ the challenges arising while‌ designing such predictive models‌‌ in the context of rare and pharmacoresistant epilepsies.‌ Specifically, we investigate how‌ to address current methodological‌‌ limitations in drug response and polypharmacy effects by‌ leveraging graph-based modeling. We‌ aim to employ GNN‌‌ models enhanced with (i) self-supervision to address label‌ scarcity, (ii) transfer learning‌ to cope with the‌‌ small number of patients due to the rare‌ nature of the diseases‌ studied, (iii) and graph‌‌ explainability to understand model predictions. This will allow‌ us to analyze patient‌ similarity and drug-disease networks‌‌ towards unraveling hidden relationships among drugs, epilepsies, and‌ patients; and ultimately discovering‌ patient phenotyping clusters for‌‌ precision medicine to improve therapies' efficacy.

4.4.5 Graph‌ inference for time series‌ analysis in complex dynamical‌‌ systems

Participants:Emilie Chouzenoux (Collaboration: V. Elvira, Univ.‌ Edinburgh, UK)

Studying the‌ causal inter-dependencies in complex‌‌ dynamical systems is a critical challenge. We explore‌ sparse graphical models to‌ gain insight through observational‌‌ causal discovery. We revisit Granger causality under a‌ graphical perspective of state-space‌ models. We investigate differentiable‌‌ particle filtering coupled with proximal methods in 35‌11 for estimating parameters‌ in the state equation‌‌ of a polynomial state-space model under sparse priors,‌ emphasizing both causal and‌ correlation relationships among time‌‌ series samples.

4.5 Other biomedical applications

4.5.1 Visual‌ Language Foundation Models for‌ medical image diagnosis

Participants‌‌: Maria Vakalopoulou

This research axis investigates visual–language‌ foundation models for medical‌ image diagnosis with a‌‌ strong emphasis on robustness, fairness, and performance. It‌ aims to study and‌ design multimodal models that‌‌ maintain reliable diagnostic accuracy under data shifts, noise,‌ and limited annotations, while‌ ensuring equitable performance across‌‌ patient populations, imaging devices, and clinical settings. The‌ axis explores modality biases‌ for state of the‌‌ art VLMs, uncertainty estimation, and rigorous evaluation protocols‌ to enhance generalization, clinical‌ safety, and trustworthiness, ultimately‌‌ enabling high-performing and fair AI-assisted diagnostic systems.

4.5.2‌ Independant vector analysis for‌ fMRI data processing

Participants‌‌: Emilie Chouzenoux , Clement Cosserat (Collaboration: Univ.‌ Baltimore ; GeorgiaTech)

Analyzing‌ multi-subject functional magnetic resonance‌‌ imaging (fMRI) data requires methods that can jointly‌ capture shared and individual‌ patterns of brain activity‌‌ across participants. Through a collaboration with Prof. Tulay‌ Adali (Univ. Baltimore), and‌ medical doctors from GeorgiaTech,‌‌ we explore joint blind source separation (JBSS) techniques,‌ and in particular, independent‌ vector analysis (IVA), to‌‌ model dependencies across subjects while identifying distinct functional‌ networks 34.

4.5.3‌ Imaging biomarkers and characterization‌‌ for chronic lung diseases

Participants: Othmane Laousy‌ , Maria Vakalopoulou (Collaboration:‌ S. Christodoulidis, G. Chassagnon,‌‌ M.-P. Revel, APHP ; N. Paragios, Therapanacea)

Diagnosis‌ and staging of lung‌ diseases is a major‌‌ challenge for both patient care and approval of‌ new treatments. Among imaging‌ techniques, computed tomography (CT)‌‌ is the gold standard‌ for in vivo morphological assessment of lung anatomy‌ currently offering the highest spatial resolution in lung‌ diseases. Although CT is widely used its optimal‌ use in clinical practice and as an endpoint‌ in clinical trials remains controversial. Our goal in‌ the PhD thesis of Othmane Laousy , is‌ to provide automatic and accurate tools that could‌ help clinicians with their everyday practice.

4.5.4 AI‌ for small bowel obstruction diagnosis

Participants:Emilie Chouzenoux‌ , Maxence Gélard, Jean-Christophe Pesquet , Quentin Vanderbecq‌ (Collaboration:M. Zins, Hôpital Saint Joseph ; M. Wagner,‌ LIB, Sorbonne Univ.)

Small bowel obstruction (SBO) is‌ a common nontraumatic surgical emergency. All guidelines recommend‌ computed tomography (CT) as the first-line imaging technique‌ for patients with suspected mechanical SBO with a‌ four-fold goal: (i) to confirm or refute the‌ diagnosis of SBO and, when SBO is present,‌ (ii) to locate the site of the obstruction,‌ that is, the transition zone (iii) to identify‌ the cause, and (iv) to look for complications‌ such as strangulation or perforation. Identifying SBO and‌ differentiating its causes (e.g., open-loop and closed-loop mechanisms)‌ is time-consuming and subject to inter-observer and intra-observer‌ variability.

The aim of this collaborative project between‌ Inria Saclay OPIS, Hôpital St Joseph, and LIB,‌ Sorbonne University, is to investigate AI approaches for‌ a guided SBO diagnosis from 3D CT scans.‌

4.5.5 A generative model for heart left ventricle‌ perfusion analysis

Participant:Hugues Talbot (Collaboration: L. Najman,‌ ESIEE Paris ; I. Vignon-Clementel, REO Team leader,‌ Inria ; C. Taylor, Heartflow Inc.)

Cardio-vascular diseases‌ continue to be the leading cause of mortality‌ in the world. Understanding these diseases is a‌ current, challenging and essential research project. The leading‌ cause of heart malfunction are stenoses causing ischemia‌ in the coronary vessels. Current CT and MRI‌ technology can assess coronary diseases but are typically‌ invasive, requiring risky catheterization and renal-toxic contrast agents‌ injection. In collaboration with the REO team headed‌ by Irène Vignon-Clementel, and Heartflow, a US based‌ company, we have in the past contributed to‌ Heartflow's major product, that replaces these physical exams‌ with image-based exams only, limiting the use of‌ contrast agents and in the cases that do‌ not require a stent insertion, eliminating catheterisation. Heartflow‌ is current the market leader in non-invasive coronary‌ exams and the owner of most of the‌ relevant IP in this domain.

Unfortunately, current imaging‌ technology is unable to assess coronary disease along‌ the full length of coronary vessels. CT is‌ limited to a resolution of about 1mm, whereas‌ coronary vessels can be much smaller, down to‌ about 10 micrometers in diameter. To assess blood‌ vessel health down to the smallest sizes, blood‌ perfusion imaging technique throughout the heart muscle must‌ be used instead. Perfusion imaging with PET or‌ a Gamma camera, the current gold standard, is‌ an invasive technology requiring the use of radioactive‌ tracers. To avoid using these, a lower quality‌ estimate of perfusion can be achieved using some ToF or injected gated‌ MRI modalities.

We have‌ investigated patient-specific vessel generation‌‌ models together with porous model simulations in order‌ to propose a direct‌ model of perfusion imaging,‌‌ based on the known patient data, computer flow‌ dynamic simulations as well‌ as experimental data consistent‌‌ with known vessel and heart muscle physiology. The‌ objective of this work‌ is to both to‌‌ provide a useful, complex forward model of perfusion‌ image generation, and to‌ solve the inverse problem‌‌ of locating and assessing coronary diseases given a‌ perfusion exam, even though‌ the affected vessels may‌‌ be too small to be imaged directly.

Continuing‌ on our work from‌ the period 2015-2019, this‌‌ year we proposed a functional myocardial perfusion model‌ consisting of the CT-derived‌ segmented coronary vessels, a‌‌ simulated vessel tree consisting of several thousands of‌ terminal vessels, filling the‌ myocardium in a patient-specific‌‌ way, consistent with physiology data, physics-based and empirically-observed‌ vessel growth rules, and‌ a porous medium. We‌‌ produced and validated a CFD code capable of‌ simulating blood flow in‌ all three coupled compartments,‌‌ which allows us to simulate perfusion realistically.

5‌ Social and environmental responsibility‌

5.1 Footprint of research‌‌ activities

The research carried out in OPIS aims‌ at developing advanced techniques‌ in the domain of‌‌ data science for precision medicine. One of the‌ main features of this‌ research is to ensure‌‌ that the proposed methods are not only efficient,‌ but also grounded on‌ sound mathematical foundations inherited‌‌ from the areas of optimization and fixed point‌ algorithms. In the biomedical‌ domain, it appears indeed‌‌ mandatory to guarantee the reliability and the explainability‌ of the proposed approaches‌ in their use by‌‌ medical doctors or producers of medical imaging devices.‌

5.2 Impact of research‌ results

OPIS participates in‌‌ the design of innovative products developed by big‌ companies working in the‌ domain of medical imaging‌‌ (GE Healthcare and Essilor) and several startups. Various‌ application fields are targeted‌ (breast cancer detection, surgical‌‌ radiology, interventional surgery, coronary disease monitoring, vision correction,‌ ...).

The methodological contributions‌ of OPIS are far‌‌ reaching, with impact going further the field of‌ medical imaging. OPIS transfers‌ its expertise in artificial‌‌ intelligence, image processing, and optimization through collaboration with‌ major industrial partners such‌ as SNCF, Schneider Electrics,‌‌ IFPEN, and Thales. The transfer activity typically goes‌ through CIFRE PhD contracts‌ or more dedicated partnerships.‌‌

In addition, OPIS has active collaborations with several‌ hospitals, particularly Institut Gustave‌ Roussy and public hospitals‌‌ from APHP in Paris. The purpose of these‌ collaborations is to develop‌ artificial intelligence tools aiding‌‌ medical doctors in their practice. A large part‌ of this research activity‌ is oriented toward fighting‌‌ against cancer using different kinds of data (CT‌ scans, MRI, genomic data,‌ histopathology images,...). OPIS was‌‌ also involved in several projects for helping to‌ better diagnose and cure‌ COVID-19 infection.

6 Highlights‌‌ of the year

6.1 Awards

Emilie Chouzenoux received‌ the EURASIP Early Career‌ Award in 2025 for‌‌ exceptional contributions to optimization‌ for signal and image processing with successful application‌ to medical imaging.
Jean-Christophe Pesquet was awarded an‌ Otto Monsted visiting professorship at Danish Technical University.‌
Maria Vakalopoulou has received the best poster award‌ at the FAIMI Workshop of MICCAI'2025 for the‌ paper 43.
Antonio Silveti-Falls got a spotlight‌ paper at ICML'2025 40, and an oral‌ paper at NEURIPS'2025 41.

7 Latest software‌ developments, platforms, open data

7.1 New platforms

7.1.1‌ Prox Repository

Web site: Prox Repository

Software Family:‌ utility
Audience: universe
Evolution and maintenance: long term‌ support
Context/Role of OPIS: This website was created‌ by E. Chouzenoux and J.-C. Pesquet from OPIS,‌ along with P.L. Combettes, North Carolina State University,‌ and G. Chierchia, ESIEE Paris. The maintenance is‌ made by summer interns funded by OPIS, and‌ by the authors of the website.
Duration of‌ the Development: The website was released in 2016,‌ and is maintained regularly since then.
Proximity operators‌ have become increasingly important tools as basic building‌ blocks of proximal splitting algorithms, a class of‌ algorithms that decompose complex composite convex optimization methods‌ into simple steps involving one of the functions‌ present in the model. This website provides formulas‌ for efficiently computing the proximity operator of various‌ functions, along with the associated codes in Matlab/Python‌ langages.
The codes provided are distributed under the‌ licence CeCill-B.

7.1.2 The PINK image library

Web‌ site: PINK

Software Family: utility
Audience: universe
Evolution‌ and maintenance: long term support
Context/Role of OPIS:‌ H. Talbot is among the creators of this‌ library and is still actively involved in its‌ maintenance.
Duration of the Development: This software has‌ been developed and maintained since 2011.
The PINK‌ image library is a general-purpose, open-source, portable image‌ processing library specializing in discrete geometry and mathematical‌ morphology. It is the result of several decades‌ of research in these domains and features state-of-the‌ art algorithmic implementation of both classical and leading‌ edge DG and MM operators. These include nD‌ parallel thinning and skeletonization methods and efficient hierarchical‌ morphological transforms.
This code is distributed under the‌ CeCILL license.

7.1.3 The Vivabrain AngioTK toolkit

Web‌ site: Vivabrain AngioTK toolkit

Software Family: vehicle
Audience:‌ partners
Evolution and maintenance: basic
Context/Role of OPIS:‌ H. Talbot participated to the programming of this‌ software.
Duration of the Development: This software has‌ been released in 2018.
AngioTK is a toolkit‌ supported by Kitware (the authors of VTK) for‌ the filtering, segmentation, generation and simulation of blood‌ vessels. It was started in the context of‌ the Vivabrain ANR project in 2012, but continues‌ with the same as well as new partners.‌ Applications are numerous, from the simulation and understanding‌ of perfusion (see associated theme) to the simulation‌ of realistic blood flow MRI images with associated‌ ground truth, via the generation of blood vessel‌ atlases.
This code is distributed under the Apache‌ License, Version 2.0.

7.1.4 A scientific image viewer‌

Web site: imview

Software Family: vehicle
Audience: community
Evolution and maintenance: basic‌
Context/Role of OPIS: H.‌ Talbot is the author‌‌ of this software.
Duration of the Development: This‌ software has been released‌ in 2001. His last‌‌ version has been updated in 2014.
This general-purpose‌ and cross-platform scientific image‌ viewing tool has been‌‌ part of the Debian Linux distribution since 2001.‌ This viewer is well‌ adapted to display 2D‌‌ with high-precision data images (floating-point, etc), as well‌ as 3D and hyper-spectral‌ data. It features an‌‌ interactive segmentation tool for multispectral data and is‌ scriptable.
The codes provided‌ are distributed under a‌‌ GNU General Public License version 2.0 (GPLv2).

7.1.5‌ Scion

Web site: Scion‌

Software Family: vehicle
Audience:‌‌ community
Evolution and maintenance: basic
Context/Role of OPIS:‌ This software is developed‌ by Antonio Silveti-Falls ,‌‌ in the context of a collaboration with EPFL‌ (LIONS/Cevher).
Duration of the‌ Development: This software was‌‌ developed during the year 2025.
This is a‌ Python code allowing to‌ reproduce the results of‌‌ the papers 40 and 41. It also‌ includes a general PyTorch‌ optimizer that can be‌‌ used to train neural networks.
The codes provided‌ are distributed under an‌ MIT License.
The code‌‌ also appears in the NVIDA Nemotron codebase.‌

7.1.6 ScanCovIA

Web site:‌ ScanCovIA

Software Family: vehicle‌‌
Audience: community
Evolution and maintenance: basic
Context/Role of‌ OPIS: This software is‌ the product of the‌‌ ScanCovIA collaborative project between OPIS, IGR, CentraleSupélec and‌ the start-up Owkin. Several‌ PhD students of OPIS‌‌ were involved in the programming of this software.‌
Duration of the Development:‌ This software was developed‌‌ during the year 2020.
This is a Python/Pytorch‌ code allowing to reproduce‌ the results of the‌‌ ScanCovIA project.
The codes provided are distributed under‌ a GPL v3.0 license.‌

7.1.7 FOSSIL

Web site:‌‌ FOSSIL

Software Family: vehicle
Audience: community
Evolution and‌ maintenance: basic
Duration of‌ the Development: This software‌‌ is developed as part of a collaboration with‌ Télécom Paris, in the‌ context of Amadou Siaka‌‌ Sangare's research project and Nicola Dunou's research internship‌ (co-supervised by F. Malliaros‌ and J. Giraldo).
This‌‌ is a Python code allowing to reproduce the‌ results of the paper‌ 26.
The codes‌‌ provided are distributed under the licence GNU General‌ Public License v3.0.

7.1.8‌ PieCoN

Web site: PieCoN‌‌

Software Family: vehicle
Audience: community
Evolution and maintenance:‌ basic
Duration of the‌ Development: This software is‌‌ developed in the context of Vahan Martirosyan's PhD‌ work (co-supervised by F.‌ Malliaros, H. Talbot, and‌‌ J. Giraldo).
This is a Python code allowing‌ to reproduce the results‌ of the paper 21‌‌.
The codes provided are distributed under the‌ licence GNU General Public‌ License v3.0.

7.1.9 Cometh‌‌

Web site: Cometh

Software Family: vehicle
Audience: community‌
Evolution and maintenance: basic‌
Duration of the Development:‌‌ This software is initially developed in the context‌ of Antoine Siraudin's research‌ internship. It continued during‌‌ his ongoing PhD at RWTH Aachen University in‌ collaboration with Christopher Morris‌ (RWTH Aachen University).
This‌‌ is a Python code‌ allowing to reproduce the results of the paper‌ 27.
The codes provided are distributed under‌ the licence GNU General Public License v3.0.

7.1.10‌ Gratin

Web site: Gratin

Software Family: vehicle
Audience:‌ community
Evolution and maintenance: basic
Duration of the‌ Development: This software is developed in the context‌ of Yassine Abbahaddou's PhD work (co-supervised by F.‌ Malliaros, J. Lutzeyer, and M. Vazirgiannis, École Polytechnique),‌ in collaboration with Mohamed Aboussalah (NYU).
This is‌ a Python code allowing to reproduce the results‌ of the paper 29.
The codes provided‌ are distributed under the licence GNU General Public‌ License v3.0.

7.1.11 FairDiffRec

Web site: FairDiffRec

Software‌ Family: vehicle
Audience: community
Evolution and maintenance: basic‌
Duration of the Development: This software is developed‌ in the context of the postdoctoral position of‌ D. Malitesta, in collaboration with G. Medda, M.‌ Marras, L. Boratto (University of Cagliari, Italy) and‌ E. Purificato (JRC, Italy).
This is a Python‌ code allowing to reproduce the results of the‌ paper 38.
The codes provided are distributed‌ under the licence GNU General Public License v3.0.‌

7.1.12 ADMP-GNN

Web site: ADMP-GNN

Software Family: vehicle‌
Audience: community
Evolution and maintenance: basic
Duration of‌ the Development: This software is developed in the‌ context of Yassine Abbahaddou's PhD work (co-supervised by‌ F. Malliaros, J. Lutzeyer, and M. Vazirgiannis, École‌ Polytechnique).
This is a Python code allowing to‌ reproduce the results of the paper 30.‌
The codes provided are distributed under the licence‌ GNU General Public License v3.0.

7.1.13 TRIGON

Web‌ site: TRIGON

Software Family: vehicle
Audience: community
Evolution‌ and maintenance: basic
Duration of the Development: This‌ software is developed as part of a collaboration‌ with Université Sorbonne Paris Nord (H. Attali, T.‌ Papastergiou, and N. Pernelle).
This is a Python‌ code allowing to reproduce the results of the‌ paper 32.
The codes provided are distributed‌ under the licence GNU General Public License v3.0.‌

7.1.14 COSIMO

Web site: COSIMO

Software Family: vehicle‌
Audience: community
Evolution and maintenance: basic
Duration of‌ the Development: This software is developed as part‌ of a collaboration with Télécom Paris, in the‌ context of the postdoctoral position of A. Einizade‌ (co-supervised by F. Malliaros and J. Giraldo), in‌ collaboration with D. Thanou (EPFL).
This is a‌ Python code allowing to reproduce the results of‌ the paper 36.
The codes provided are‌ distributed under the licence GNU General Public License‌ v3.0.

8 New results

8.1 Deep unrolled algorithms‌ for inverse problems in image and signal processing‌

Participants:Emilie Chouzenoux , Jean-Christophe Pesquet , Clement‌ Cosserat , Gaspard Blaise (Collaborations: T. Adali, Univ.‌ Baltimore, USA ; C. Delle Valle, Smartway)

While‌ model-based iterative methods can be used for solving‌ inverse problems arising in image processing, their practicability‌ might be limited due to tedious parameterization and‌ slow convergence. In addition, inadequate solutions can be‌ obtained when the retained priors do not perfectly‌ fit the solution space. Deep learning methods offer an alternative approach that‌ is fast, leverages information‌ from large data sets,‌‌ and thus can reach high reconstruction quality. However,‌ these methods usually rely‌ on black boxes not‌‌ accounting for the physics of the imaging system,‌ and their lack of‌ interpretability is often deplored.‌‌ At the crossroads of both methods, unfolded deep‌ learning techniques have been‌ recently proposed. They incorporate‌‌ the physics of the model and iterative optimization‌ algorithms into a neural‌ network design, leading to‌‌ superior performance in various applications.

In 9,‌ we question the robustness‌ of an unrolled neural‌‌ network architecture designed to solve inverse problems where‌ the degradation operator is‌ linear and known. This‌‌ architecture is constructed by unrolling a forward-backward algorithm‌ derived from the minimization‌ of an objective function‌‌ that combines a data-fidelity term, a Tikhonov-type regularization‌ term, and a potentially‌ nonsmooth convex penalty. The‌‌ robustness of this inversion method to input perturbations‌ is analyzed theoretically.

In‌ 33, we propose‌‌ U-PALM-IVA-G, an unrolled implementation of the proximal alternating‌ algorithm PALM-IVA-G 10.‌ The approach allows to‌‌ perform Gaussian independent vector analysis (IVA-G), a blind‌ source separation method which‌ models source datasets as‌‌ independent Gaussian vectors and estimates both precision and‌ demixing matrices.

8.2 Unrolled‌ Generalized EM for Transductive‌‌ Few-Shot Learning

Participants:Aymen Sadraoui , Mounir Kaaniche‌ , Jean-Christophe Pesquet ,‌ Long Zhou (Collaboration: Ismail‌‌ Ben Ayed, ETS Montréal, Canada)

Few-Shot Learning (FSL)‌ has recently attracted wide‌ attention within the computer‌‌ vision community. In this respect, transductive approaches, in‌ which inference is performed‌ jointly on a batch‌‌ of query samples, introduce key hyper-parameters that control‌ the prediction statistics of‌ the test batches, such‌‌ as the level of class balance, affecting performances‌ significantly. Such hyper-parameters are‌ empirically grid-searched over validation‌‌ data, and their configurations may vary substantially with‌ the target dataset and‌ pre-training model, making such‌‌ empirical searches both sub-optimal and computationally intractable. To‌ address these challenges, we‌ introduced the unrolling paradigm,‌‌ also referred to as 'learning to optimize', in‌ the context of few-shot‌ learning, thereby learning efficiently‌‌ and effectively a set of optimized hyperparameters 46‌. Specifically, we unroll‌ a generalization of the‌‌ ubiquitous Expectation-Maximization (EM) optimizer into a neural network‌ architecture, mapping each of‌ its iterates to a‌‌ layer and learning a set of key hyper-parameters‌ over validation data. Our‌ unrolling approach covers various‌‌ statistical feature distributions and pre-training paradigms, including recent‌ foundational vision-language models and‌ standard vision-only classifiers. We‌‌ report comprehensive experiments, which cover a breadth of‌ fine-grained downstream image classification‌ tasks, showing significant gains‌‌ brought by the proposed unrolled EM algorithm over‌ iterative variants.

8.3 Learning‌ truly monotone operators

Participants:‌‌Jean-Christophe Pesquet , Hugues Talbot (Collaboration: A. Repetti,‌ Heriot-Watt University, Edinburgh ;‌ Y. Belkouchi, GLEAMER)

We‌‌ introduce in 5 a novel approach to learning‌ monotone neural networks through‌ a newly defined penalization‌‌ loss. The proposed method is particularly effective in‌ solving classes of variational‌ problems, specifically monotone inclusion‌‌ problems, commonly encountered in‌ image processing tasks. The Forward-Backward-Forward (FBF) algorithm is‌ employed to address these problems, offering a solution‌ even when the Lipschitz constant of the neural‌ network is unknown. Notably, the FBF algorithm provides‌ convergence guarantees under the condition that the learned‌ operator is monotone. Building on plug-and-play methodologies, our‌ objective is to apply these newly learned operators‌ to solving non-linear inverse problems. To achieve this,‌ we initially formulate the problem as a variational‌ inclusion problem. Subsequently, we train a monotone neural‌ network to approximate an operator that may not‌ inherently be monotone. Leveraging the FBF algorithm, we‌ then show simulation examples where the non-linear inverse‌ problem is successfully solved.

8.4 Majorization-Minimization algorithm and‌ application to Dirichlet maximum likelihood estimation

Participants:Jean-Christophe‌ Pesquet (Collaboration: S. Martin, Inria Lyon ; G.‌ Steild, TU Berlin ; I. Ben ayed, ETS‌ Montréal)

We propose in 20 a novel Bregman‌ descent algorithm for minimizing a convex function that‌ is expressed as the sum of a differentiable‌ part (defined over an open set) and a‌ possibly nonsmooth term. The approach, referred to as‌ the Variable Bregman Majorization-Minimization (VBMM) algorithm, extends the‌ Bregman Proximal Gradient method by allowing the Bregman‌ function used in the divergence to adaptively vary‌ at each iteration, provided it satisfies a majorizing‌ condition on the objective function. This adaptive framework‌ enables the algorithm to approximate the objective more‌ precisely at each iteration, thereby allowing for accelerated‌ convergence compared to the traditional Bregman Proximal Gradient‌ descent. We establish the convergence of the VBMM‌ algorithm to a minimizer under mild assumptions on‌ the family of metrics used. Furthermore, we introduce‌ a novel application of both the Bregman Proximal‌ Gradient method and the VBMM algorithm to the‌ estimation of the multidimensional parameters of a Dirichlet‌ distribution through the maximization of its log-likelihood. Numerical‌ experiments confirm that the VBMM algorithm outperforms existing‌ approaches in terms of convergence speed.

8.5 Multi-task‌ neural networks for lifting-based image coders

Participants: Tassnim‌ Dardouri, Mounir Kaaniche , Jean-Christophe Pesquet (Collaboration: Amel‌ Benazza-Benyahia, SUP'COM-Tunis)

Motivated by the several advantages of‌ lifting-based representations and the promising results shown by‌ our recent FCNN-LS-based coding methods, the use of‌ neural networks in lifting-based image coding systems has‌ been further investigated in 12. While considering‌ a popular non-separable lifting structure that relies on‌ three prediction stages and an update stage, we‌ propose to perform the different involved lifting steps‌ by using CNN models to better capture the‌ local structure of the input image. Most importantly,‌ unlike previous works where different neural network models‌ are employed to carry out the LS-based decomposition‌ at a given resolution level, a new multi-task‌ CNN architecture is developed. The proposed architecture aims‌ to exploit the similarities between the second and‌ third prediction steps and perform their learning in‌ a joint manner. The experimental results, obtained with‌ different standard image datasets, have shown the good‌ performance of the proposed approach compared to the state-of-the-art methods, and more‌ specifically, the recent neural‌ networks-based lifting schemes.

8.6‌‌ Training Neural Networks with non-Euclidean geometry and spectral‌ constraints

Participants:Antonio Silveti-Falls‌ (Collaboration: V. Cevher, K.‌‌ Antonakopoulos, W. Xie, T. Pethick, Z. Zhu, M.‌ Erdogan, EPFL ; L.‌ Chennuru-Vankadara, UCL Gatsby)

Scaling‌‌ the training of deep neural networks remains a‌ central challenge, with standard‌ optimization methods often requiring‌‌ extensive hyperparameter tuning as model size increases. In‌ this line of work,‌ we investigate training methods‌‌ based on non-Euclidean geometry and spectral norm constraints,‌ which offer principled approaches‌ to controlling weight dynamics‌‌ during optimization. Our framework is modular and architecture-agnostic,‌ applying uniformly to MLPs,‌ convolutional networks, transformers, and‌‌ other layer types through norm-constrained linear minimization oracles‌ (LMOs). In 40,‌ we introduced the Scion‌‌ optimizer, which leverages this geometric perspective to achieve‌ significant speedups on language‌ model training while remaining‌‌ memory-efficient. We extended this framework in 41 which‌ introduces Clipped Scion—a hybrid‌ method combining steepest descent‌‌ with conditional gradient approaches under a generalized smoothness‌ condition, with principled weight‌ decay and order-optimal convergence‌‌ rates in the stochastic setting. Finally, 51 provides‌ a unified survey connecting‌ these optimization techniques to‌‌ neural network architectures, demonstrating how spectral geometry emerges‌ naturally from feature propagation‌ analysis and enables hyperparameter‌‌ transfer across model scales.

8.7 Variational morphological operators‌ on graphs

Participants:Hugues‌ Talbot , Tristan Portugues‌‌ , Antonio Silveti-Falls , Jean-Christophe Pesquet , Miguel‌ Amorim

The work 31‌ explores mathematical morphology applied‌‌ to irregular domains like graphs. We develop a‌ numerical implementation of variational‌ morphological operators (like erosion‌‌ and dilation) that can function on complex network‌ structures. This allows for‌ advanced geometric analysis and‌‌ filtering on non-Euclidean data, which is increasingly important‌ for 3D mesh processing.‌ This works paves the‌‌ way to variational image processing with morphological operators,‌ such as solving inverse‌ problems on quantized data,‌‌ or with known contrast constraints.

8.8 Segmentation of‌ fiber bundles in tomography‌

Participants:Hafsa El Herichi‌‌ , Hugues Talbot (Collaboration: S. Roux, ENS Paris-Saclay)‌

This study focuses on‌ composite materials and the‌‌ difficulty of segmenting "torons" (fiber bundles) in X-ray‌ tomographic images in the‌ context of reinforcement materials‌‌ for civil aeronautics, particularly in large commercial jet‌ engines. We utilize an‌ a priori shape analysis‌‌ approach to guide the segmentation process. By incorporating‌ known geometric constraints of‌ the fiber structures into‌‌ the algorithm, we achieve much higher accuracy in‌ identifying individual bundles within‌ dense, complex composite mateials,‌‌ at a scale never attempted before. In 37‌, we propose an‌ inexpensive, explainable learning-based methods‌‌ not requiring GPUs or large amounts of memory,‌ allowing us to process‌ a significant volume of‌‌ the data at once.

8.9 Convergence analysis in‌ stochastic optimization

Participants:‌ Emilie Chouzenoux , Jean-Baptiste‌‌ Fest (Collaboration: A. Repetti, Heriot-Watt Univ., Edinburgh)

Asymptotic‌ analysis of generic stochastic‌ algorithms often relies on‌‌ descent conditions. In a convex setting, some technical‌ shortcuts can be considered‌ to establish asymptotic convergence‌‌ guarantees of the associated‌ scheme. However, in a non-convex setting, obtaining similar‌ guarantees is usually more complicated, and relies on‌ the use of the Kurdyka-Lojasiewicz (KL) property. In‌ 15, we propose a new framework for‌ using the KL property in a non-convex stochastic‌ setting based on conditioning theory.

8.10 Divergence minimization‌ in statistical inference

Participants:Emilie Chouzenoux , Thomas‌ Guilmeau (Collaboration: V. Elvira, N. Branchini, Univ. Edinburgh)‌

A wide class of problems in statistical inference,‌ including proposal adaptation in Monte-Carlo, maximum likelihood estimation,‌ and variational approximation, read as the minimization of‌ a divergence over a set of parametric distributions.‌ We investigate the resolution of such problems, with‌ modern tools of convex analysis. In 16,‌ we study the variational inference problem of minimizing‌ a regularized Rényi divergence over an exponential family,‌ and propose a relaxed moment-matching algorithm, which includes‌ a proximal-like step. Using the information-geometric link between‌ Bregman divergences and the Kullback-Leibler divergence, this algorithm‌ is shown to be equivalent to a Bregman‌ proximal gradient algorithm. This novel perspective allows us‌ to exploit the geometry of our approximate model‌ while using stochastic black-box updates. We use this‌ point of view to prove strong convergence guarantees‌ including monotonic decrease of the objective, convergence to‌ a stationary point or to the minimizer, and‌ convergence rates.

8.11 Fast schemes for adaptive importance‌ sampling

Participants:Emilie Chouzenoux (Collaboration: V. Elvira, University‌ of Edinburgh, UK ; O. D. Akyildiz, Imperial‌ College London, UK)

Adaptive importance sampling (AIS) methods‌ are increasingly used for the approximation of distributions‌ and related intractable integrals in the context of‌ Bayesian inference. In 14, we propose a‌ proximal Newton adaptive importance sampler for the estimation‌ of expectations with respect to non-smooth target distributions.‌ We implement a scaled Newton proximal gradient method‌ to adapt the proposal distributions, enabling efficient and‌ optimized moves even when the target distribution lacks‌ differentiability.

8.12 Computational approaches for drug discovery

Participant:‌Emilie Chouzenoux (Collaboration: S. Chatterjee, K. Kumar, S.‌ Jain, A. Majumdar, IIIT Delhi)

The discovery of‌ drug-target interactions (DTIs) is a very promising area‌ of research with great potential. The accurate identification‌ of reliable interactions among drugs and proteins via‌ computational methods, which typically leverage heterogeneous information retrieved‌ from diverse data sources, can boost the development‌ of effective pharmaceuticals. We recently focused on computational‌ models for repurposing drugs with the potential for‌ treating drug resistant bacterial infections. In 8,‌ we have developed a new algorithm for general-purpose‌ drug repositioning based on a matrix completion framework‌ on graphs. Our probabilistic approach combines deep matrix‌ factorization with graph learning to achieve precise drug‌ repurposing on three antimicrobial resistance case studies.

8.13‌ Analysis of pathology whole slide images with spatial‌ context

Participants:Loïc Le Bescond , Hugues Talbot‌ (Collaboration: M. Lerousseau, Institut Curie ; F. André,‌ Gustave-Roussy)

The computer analysis of Whole Slide Images‌ (WSI) is becoming increasingly prevalent in pathology-based diagnosis,‌ although it presents considerable challenges due to the voluminous nature of the‌ data. To address this‌ issue, Multiple Instance Learning‌‌ (MIL) has emerged as a viable approach that‌ involves partitioning WSI into‌ tiles for processing. Nevertheless,‌‌ conventional MIL methodologies inadequately capture the essential spatial‌ context between tiles, which‌ is imperative for accurate‌‌ diagnosis across various diseases. In this work 49‌, we present a‌ novel framework, SparseXceptionMIL (SparseXMIL),‌‌ aiming to enhance the modeling of spatial interactions‌ within WSI data by‌ introducing a multi-dimensional sparse‌‌ image representation and a novel pooling operator. This‌ operator integrates sparse convolutions‌ within the Xception architecture.‌‌ It enables effective spatial information processing across multiple‌ scales. Empirical evaluations conducted‌ on various classification tasks,‌‌ encompassing subtyping for breast and lung carcinomas and‌ predicting abnormalities in the‌ DNA damage response in‌‌ breast cancer WSI, consistently demonstrate the superiority of‌ our approach over benchmark‌ methods. These results underscore‌‌ the potential of sparse convolutional architectures to improve‌ WSI classification.

8.14 Robust‌ automatic crater detection at‌‌ all latitudes on Mars with Deep-learning

Participants:Hugues‌ Talbot (Collaboration: L. Martinez,‌ F. Andrieu, F. Schmidt,‌‌ Geoscience Paris-Saclay)

Understanding the distribution and characteristics of‌ impact craters on planetary‌ surfaces allows researchers to‌‌ unravel geological processes and the evolution of celestial‌ bodies. Several machine learning‌ and AI-based approaches have‌‌ been proposed to detect craters on planetary surface‌ images automatically. However, designing‌ a robust tool for‌‌ an entire complex planet such as Mars, is‌ still an open problem.‌ Our work 17 presents‌‌ a novel approach using the Faster Region-based Convolutional‌ Neural Network (Faster R-CNN)‌ for such a detection.‌‌ The proposed method involves the pre-processing, training and‌ crater detection steps, which‌ are especially designed for‌‌ robustness regarding latitude and complex geomorphological features. The‌ objectives of this studies‌ are to (i) be‌‌ robust at all latitudes and (ii) for >=‌ 1 km diameter crater‌ sizes. (iii) To propose‌‌ an open-source and re-usable algorithm that (iv) only‌ needs an image to‌ run. Extensive experiments on‌‌ high-resolution planetary imagery demonstrate excellent performances with an‌ average precision mAP >‌ 0.82 with an intersection‌‌ over union criterion IoU $\geq$ 0.5, irrespective of‌ crater scale. For mid‌ and high latitudes (higher‌‌ than 48o north and south), performance decreases‌ down to mAP $\sim‌$ 0.7, which is still‌‌ better than the current state of the art.‌ Loss of performance is‌ mostly due to strong‌‌ shadowing effects. Our results also highlight the versatility‌ and potential of our‌ robust model for automating‌‌ the analysis of craters across different celestial bodies.‌ The automated crater detection‌ tool presented in this‌‌ article is publicly available as open-source and holds‌ great promise for future‌ scientific research of space‌‌ exploration missions. Published in Planetary and Space Science,‌ this paper details a‌ deep-learning system for the‌‌ automatic detection of impact craters across the Martian‌ surface. The model is‌ specifically designed to be‌‌ robust across all latitudes, handling variations in lighting,‌ terrain, and dust cover.‌ This tool is vital‌‌ for planetary scientists who‌ use crater counts to estimate the age of‌ planetary surfaces and understand the geological history of‌ Mars.

8.15 Expectation-Maximization for time series modeling and‌ inference

Participant:Emilie Chouzenoux (Collaboration: B. Cox, V.‌ Elvira, Univ. Edinburgh)

Modeling and inference with multivariate‌ sequences is central in a number of signal‌ processing applications such as acoustics, social network analysis,‌ biomedical, and finance, to name a few. The‌ linear-Gaussian state-space model is a common way to‌ describe a time series through the evolution of‌ a hidden state, with the advantage of presenting‌ a simple inference procedure due to the celebrated‌ Kalman filter. A fundamental question when analyzing multivariate‌ sequences is the search for relationships between their‌ entries (or the modeled hidden states), especially when‌ the inherent structure is a non-fully connected graph.‌ In such context, graphical modeling combined with parsimony‌ constraints allows to limit the proliferation of parameters‌ and enables a compact data representation which is‌ easier to interpret by the experts.

We recently‌ introduced a novel perspective by relating this matrix‌ to the adjacency matrix of a directed graph,‌ also interpreted as the causal relationship among state‌ dimensions in the Granger-causality sense. Under this perspective,‌ in 1135, we propose GraphGrad, a‌ fully automatic approach for obtaining sparse estimates of‌ the state interactions of a non-linear state-space model‌ via a polynomial approximation. This novel methodology unveils‌ the latent structure of the data-generating process, allowing‌ us to infer both the structure and value‌ of a rich and efficient parameterisation of a‌ general state-space model. Our method utilises a differentiable‌ particle filter, combined with suitable proximal updates, to‌ estimate the model parameters.

8.16 Interpretable ensembling rules‌

Participants:Emilie Chouzenoux , Jean-Christophe Pesquet , Nguyen‌ Vu (Collaboration: I. Ben-Ayed, ETS Montréal, Canada)

Ensemble‌ learning leverages multiple models (i.e., weak learners) on‌ a common machine learning task to enhance prediction‌ performance. Basic ensembling approaches average the weak learners‌ outputs, while more sophisticated ones stack a machine‌ learning model in between the weak learners outputs‌ and the final prediction. The work 28 fuses‌ both aforementioned frameworks. We introduce an aggregated f-average‌ (AFA) shallow neural network which models and combines‌ different types of averages to perform an optimal‌ aggregation of the weak learners predictions. We emphasize‌ its interpretable architecture and simple training strategy, and‌ illustrate its good performance on the problem of‌ few-shot class incremental learning.

8.17 Boosting radiologist performance‌ using machine learning, the usecase of parotid tumours‌ diagnosis

Participants:Emilie Chouzenoux , Arnaud Quillent (Collaboration:‌ Institut Gustave Roussy)

In the work 4,‌ we develop a machine learning algorithm based on‌ magnetic resonance images characteristics to automatically classify parotid‌ gland tumours. We then compare its results with‌ the diagnoses of junior and senior radiologists in‌ order to evaluate its utility in routine practice.‌ While automatic algorithms applied to parotid tumours classification‌ have been developed in the past, we believe‌ that our study is one of the first to leverage four different‌ MRI sequences and propose‌ a comparison with clinicians.‌‌

8.18 Iterative optimization for independent vector analysis

Participants:‌Emilie Chouzenoux , Jean-Christophe‌ Pesquet , Clement Cosserat‌‌ (Collaboration: T. Adali, Univ. Baltimore)

Independent vector analysis‌ (IVA) is an attractive‌ solution to address the‌‌ problem of joint blind source separation (JBSS), that‌ is, the simultaneous extraction‌ of latent sources from‌‌ several datasets implicitly sharing some information. Among IVA‌ approaches, we focus in‌ 10 on the celebrated‌‌ IVA-G model, that describes observed data through the‌ mixing of independent Gaussian‌ source vectors across the‌‌ datasets. We formulate a cost function whose mathematical‌ properties enable the use‌ of a proximal alternating‌‌ algorithm based on closed form operators with proved‌ convergence to a critical‌ point. In 34,‌‌ we present a comparative study of several contrained‌ IVA and regression-based IVA‌ methods methods to assess‌‌ their capacity for identifying schizophrenia-related biomarkers using real‌ fMRI data from subjects‌ with schizophrenia and healthy‌‌ controls. Our results demonstrate that both constrained family‌ of methods effectively extract‌ meaningful biomarkers while the‌‌ latter achieve comparable performance at substantially reduced computational‌ cost.

8.19 Proximal algorithm‌ for joint blood flow‌‌ computation and tissue motion compensation in Doppler ultrafast‌ ultrasound imaging

Participants:Nora‌ Ouzir , Jean-Christophe Pesquet‌‌ (Collaboration: IRIT)

Accurate tissue-clutter rejection and blood flow‌ estimation remain challenging in‌ ultrasound imaging. Traditionally, this‌‌ estimation is performed by assuming static tissues. Only‌ a few preprocessing techniques‌ attempt to deal with‌‌ the more realistic but challenging scenario where the‌ tissues are moving. The‌ paper 24 tackles this‌‌ scenario and presents a novel method for computing‌ blood flow from moving‌ tissues in ultrafast ultrasound‌‌ imaging. The proposed computational ultrasound imaging method solves‌ a global inverse problem‌ that jointly computes blood‌‌ flow, tissues, and their motions.

8.20 A Fused‌ Gromov-Wasserstein Approach to Subgraph‌ Contrastive Learning

Participants:Fragkiskos‌‌ Malliaros (Collaboration: Amadou Siaka Sangare, Jhony H. Giraldo‌ , Télécom Paris ;‌ and Nicolas Dunou, Université‌‌ Paris Dauphine-PSL).

Self-supervised learning has become a key‌ method for training deep‌ learning models when labeled‌‌ data is scarce or unavailable. While graph machine‌ learning holds great promise‌ across various domains, the‌‌ design of effective pretext tasks for self-supervised graph‌ representation learning remains challenging.‌ Contrastive learning, a popular‌‌ approach in graph self-supervised learning, leverages positive and‌ negative pairs to compute‌ a contrastive loss function.‌‌ However, current graph contrastive learning methods often struggle‌ to fully use structural‌ patterns and node similarities.‌‌ To address these issues, we present a new‌ method called Fused Gromov‌ Wasserstein Subgraph Contrastive Learning‌‌ (FOSSIL) 26. Our model integrates node-level and‌ subgraph-level contrastive learning, seamlessly‌ combining a standard node-level‌‌ contrastive loss with the Fused Gromov-Wasserstein distance. This‌ combination helps our method‌ capture both node features‌‌ and graph structure together. Importantly, our approach works‌ well with both homophilic‌ and heterophilic graphs and‌‌ can dynamically create views for generating positive and‌ negative pairs.

8.21 Piecewise‌ Constant Spectral Graph Neural‌‌ Network

Participants:Vahan Martirosyan‌ and Fragkiskos Malliaros (Collaboration: Jhony H. Giraldo, Télécom‌ Paris).

Graph Neural Networks (GNNs) have achieved significant‌ success across various domains by leveraging graph structures‌ in data. Existing spectral GNNs, which use low-degree‌ polynomial filters to capture graph spectral properties, may‌ not fully identify the graph's spectral characteristics because‌ of the polynomial's small degree. However, increasing the‌ polynomial degree is computationally expensive and beyond certain‌ thresholds leads to performance plateaus or degradation. In‌ this paper, we introduce the Piecewise Constant Spectral‌ Graph Neural Network(PieCoN) to address these challenges 21‌. PieCoN combines constant spectral filters with polynomial‌ filters to provide a more flexible way to‌ leverage the graph structure. By adaptively partitioning the‌ spectrum into intervals, our approach increases the range‌ of spectral properties that can be effectively learned.‌ Experiments on nine benchmark datasets, including both homophilic‌ and heterophilic graphs, demonstrate that PieCoN is particularly‌ effective on heterophilic datasets, highlighting its potential for‌ a wide range of applications.

8.22 Cometh: A‌ Continuous-time Discrete-state Graph Diffusion Model

Participants:Fragkiskos Malliaros‌ (Collaboration: Antoine Siraudin and Christopher Morris, RWTH Aachen‌ University).

Discrete-state denoising diffusion models led to state-of-the-art‌ performance in graph generation, especially in the molecular‌ domain. Recently, they have been transposed to continuous‌ time, allowing more flexibility in the reverse process‌ and a better trade-off between sampling efficiency and‌ quality. Here, to leverage the benefits of both‌ approaches, we propose Cometh, a continuous-time discrete-state graph‌ diffusion model, tailored to the specificities of graph‌ data 27. In addition, we also successfully‌ replaced the set of structural encodings previously used‌ in the discrete graph diffusion model with a‌ single random-walk-based encoding, providing a simple and principled‌ way to boost the model's expressive power. Empirically,‌ we show that integrating continuous time leads to‌ significant improvements across various metrics over state-of-the-art discrete-state‌ diffusion models on a large set of molecular‌ and non-molecular benchmark datasets.

8.23 Graph Neural Network‌ Generalization With Gaussian Mixture Model Based Augmentation

Participants:‌Fragkiskos Malliaros (Collaboration: Yassine Abbahaddou, Johannes F. Lutzeyer,‌ Michalis Vazirgianni, École Polytechnique ; and Amine M.‌ Aboussalah, Joint Research Center).

Graph Neural Networks (GNNs)‌ have shown great promise in tasks like node‌ and graph classification, but they often struggle to‌ generalize, particularly to unseen or out-of-distribution (OOD) data.‌ These challenges are exacerbated when training data is‌ limited in size or diversity. To address these‌ issues, we introduce a theoretical framework using Rademacher‌ complexity to compute a regret bound on the‌ generalization error and then characterize the effect of‌ data augmentation 29. This framework informs the‌ design of GRATIN, an efficient graph data augmentation‌ algorithm leveraging the capability of Gaussian Mixture Models‌ (GMMs) to approximate any distribution. Our approach not‌ only outperforms existing augmentation techniques in terms of‌ generalization but also offers improved time complexity, making‌ it highly suitable for real-world applications.

8.24 How‌ Fair is Your Diffusion Recommender Model?

Participants:Daniele‌ Malitesta and Fragkiskos Malliaros (Collaboration: Giacomo Medda, Mirko Marras, Ludovico Boratto, University‌ of Cagliari ; Erasmo‌ Purificato, NYU).

Diffusion-based learning‌‌ has settled as a rising paradigm in generative‌ recommendation, outperforming traditional approaches‌ built upon variational autoencoders‌‌ and generative adversarial networks. Despite their effectiveness, concerns‌ have been raised that‌ diffusion models - widely‌‌ adopted in other machine-learning domains - could potentially‌ lead to unfair outcomes,‌ since they are trained‌‌ to recover data distributions that often encode inherent‌ biases. Motivated by the‌ related literature, and acknowledging‌‌ the extensive discussion around bias and fairness aspects‌ in recommendation, we propose,‌ to the best of‌‌ our knowledge, the first empirical study of fairness‌ for DiffRec, chronologically the‌ pioneer technique in diffusion-based‌‌ recommendation 38. Our empirical study involves DiffRec‌ and its variant L-DiffRec,‌ tested against nine recommender‌‌ systems on two benchmarking datasets to assess recommendation‌ utility and fairness from‌ both consumer and provider‌‌ perspectives. Specifically, we first evaluate the utility and‌ fairness dimensions separately and,‌ then, within a multi-criteria‌‌ setting to investigate whether, and to what extent,‌ these approaches can achieve‌ a trade-off between the‌‌ two. While showing worrying trends in alignment with‌ the more general machine-learning‌ literature on diffusion models,‌‌ our results also indicate promising directions to address‌ the unfairness issue in‌ future work.

8.25 ADMP-GNN:‌‌ Adaptive Depth Message Passing GNN

Participants:Fragkiskos Malliaros‌ (Collaboration: Yassine Abbahaddou, Johannes‌ F. Lutzeyer, Michalis Vazirgiannis,‌‌ École Polytechnique).

Graph Neural Networks (GNNs) have proven‌ to be highly effective‌ in various graph learning‌‌ tasks. A key characteristic of GNNs is their‌ use of a fixed‌ number of message-passing steps‌‌ for all nodes in the graph, regardless of‌ each node's diverse computational‌ needs and characteristics. Through‌‌ empirical real-world data analysis, we demonstrate that the‌ optimal number of message-passing‌ layers varies for nodes‌‌ with different characteristics. This finding is further supported‌ by experiments conducted on‌ synthetic datasets. To address‌‌ this, we propose Adaptive Depth Message Passing GNN‌ (ADMP-GNN), a novel framework‌ that dynamically adjusts the‌‌ number of message passing layers for each node,‌ resulting in improved performance‌ 30. This approach‌‌ applies to any model that follows the message‌ passing scheme. We evaluate‌ ADMP-GNN on the node‌‌ classification task and observe performance improvements over baseline‌ GNN models.

8.26 Dynamic‌ Triangulation-Based Graph Rewiring for‌‌ Graph Neural Networks

Participants:Fragkiskos Malliaros (Collaboration: Hugo‌ Attali, Thomas Papastergiou, Nathalie‌ Pernelle, Université Sorbonne Paris‌‌ Nord).

Graph Neural Networks (GNNs) have emerged as‌ the leading paradigm for‌ learning over graph-structured data.‌‌ However, their performance is limited by issues inherent‌ to graph topology, most‌ notably oversquashing and oversmoothing.‌‌ Recent advances in graph rewiring aim to mitigate‌ these limitations by modifying‌ the graph topology to‌‌ promote more effective information propagation. In this work,‌ we introduce TRIGON, a‌ novel framework that constructs‌‌ enriched, non-planar triangulations by learning to select relevant‌ triangles from multiple graph‌ views 32. By‌‌ jointly optimizing triangle selection and downstream classification performance,‌ our method produces a‌ rewired graph with markedly‌‌ improved structural properties such‌ as reduced diameter, increased spectral gap, and lower‌ effective resistance compared to existing rewiring methods. Empirical‌ results demonstrate that TRIGON outperforms state-of-the-art approaches on‌ node classification tasks across a range of homophilic‌ and heterophilic benchmarks.

8.27 Dynamic Triangulation-Based Graph Rewiring‌ for Graph Neural Networks

Participants:Fragkiskos Malliaros (Collaboration:‌ Aref Einizade, Jhony H. Giraldo, Télécom Paris ;‌ Dorina Thanou, EPFL).

Simplicial complexes provide a powerful‌ framework for modeling higher-order interactions in structured data,‌ making them particularly suitable for applications such as‌ trajectory prediction and mesh processing. However, existing simplicial‌ neural networks (SNNs), whether convolutional or attention-based, rely‌ primarily on discrete filtering techniques, which can be‌ restrictive. In contrast, partial differential equations (PDEs) on‌ simplicial complexes offer a principled approach to capture‌ continuous dynamics in such structures. In this work,‌ we introduce continuous simplicial neural network (COSIMO), a‌ novel SNN architecture derived from PDEs on simplicial‌ complexes 36. We provide theoretical and experimental‌ justifications of COSIMO's stability under simplicial perturbations. Furthermore,‌ we investigate the over-smoothing phenomenon-a common issue in‌ geometric deep learning-demonstrating that COSIMO offers better control‌ over this effect than discrete SNNs. Our experiments‌ on real-world datasets demonstrate that COSIMO achieves competitive‌ performance compared to state-of-the-art SNNs in complex and‌ noisy environments.

8.28 Testing of foundation models and‌ visual language models on medical diagnosis

Participant:Maria‌ Vakalopoulou (Collaboration: MICS CentraleSupélec ; CONICET Argentina)

Clinical‌ decision-making relies on the integrated analysis of medical‌ images and the associated clinical reports. While Vision-Language‌ Models (VLMs) can offer a unified framework for‌ such tasks, they can exhibit strong biases toward‌ one modality, frequently overlooking critical visual cues in‌ favor of textual information. In this work 42‌, we introduce Selective Modality Shifting (SMS), a‌ perturbation-based approach to quantify a model’s reliance on‌ each modality in binary classification tasks. By systematically‌ swapping images or text between samples with opposing‌ labels, we expose modality-specific biases. We assess six‌ open-source VLMs-four generalist models and two fine-tuned for‌ medical data— on two medical imaging datasets with‌ distinct modalities: MIMIC-CXR (chest X-ray) and FairVLMed (scanning‌ laser ophthalmoscopy). By assessing model performance and the‌ calibration of every model in both unperturbed and‌ perturbed settings, we reveal a marked dependency on‌ text input, which persists despite the presence of‌ complementary visual information. We also perform a qualitative‌ attention-based analysis which further confirms that image content‌ is often overshadowed by text details. Our findings‌ highlight the importance of designing and evaluating multimodal‌ medical models that genuinely integrate visual and textual‌ cues, rather than relying on single-modality signals.

Moreover,‌ in 43, we extensively evaluate six widely‌ used CLIP-based models on chest X-ray classification using‌ three publicly available datasets: MIMIC-CXR, NIH-CXR14, and NEATX.‌ We assess the models fairness across six conditions‌ and patient subgroups based on age, sex, and‌ race. Additionally, we assess the robustness to shortcut‌ learning by evaluating performance on pneumothorax cases with‌ and without chest drains. Our results indicate performance gaps between patients of‌ different ages, but more‌ equitable results for the‌‌ other attributes. Moreover, all models exhibit lower performance‌ on images without chest‌ drains, suggesting reliance on‌‌ spurious correlations. We further complement the performance analysis‌ with a study of‌ the embeddings generated by‌‌ the models. While the sensitive attributes could be‌ classified from the embeddings,‌ we do not see‌‌ such patterns using PCA, showing the limitations of‌ these visualisation techniques when‌ assessing models.

8.29 Generative‌‌ models for longitudinal analysis of lung disease progression‌

Participant:Maria Vakalopoulou (Collaboration:‌ MICS CentraleSupélec ; AP-HP‌‌ Hopital Cochin)

Longitudinal medical image studies often involves‌ multiple scans of the‌ same patient taken at‌‌ different times, potentially with different modalities such as‌ (2D vs. 3D volumetric‌ medical imaging). In this‌‌ work 39, we propose a single diffusion-based‌ framework that can predict‌ future embeddings of imaging‌‌ data for predefined time points. Our approach uses‌ a universal vision encoder,‌ able to ingest either‌‌ 2D or 3D scans, combined with a temporal‌ transformer to fuse embeddings‌ across multiple timepoints. A‌‌ conditional latent diffusion model then produces the future‌ output in latent space‌ encoding the longitudinal information‌‌ of the patient. We challenged our method in‌ two crucial tasks involving‌ radiological imaging: (1) predicting‌‌ future pathology in the form of segmentation masks,‌ exemplified by Interstitial Lung‌ Disease (ILD) progression on‌‌ 3D chest CT scans of Systemic Sclerosis (SSc)‌ patients, and (2) generating‌ radiology reports that incorporate‌‌ prior imaging context, exemplified by longitudinal chest X-rays‌ from MIMIC-CXR. Results indicate‌ that this unified diffusion‌‌ approach outperforms existing baselines in both pixel-level forecasting‌ and report generation, highlighting‌ its versatility and effectiveness‌‌ for longitudinal medical imaging.

8.30 Inflammation Detection in‌ MRI

Participants:Hugues Talbot‌ (Collaboration: T. Aouad, Mima‌‌ Health ; A. Feydy, Université Paris-Cité, APHP)

This‌ clinical study 23 evaluates‌ a deep-learning algorithm's ability‌‌ to detect sacroiliitis (inflammation) in patients with axial‌ spondyloarthritis. By analyzing MRI‌ scans of the sacroiliac‌‌ joints, the AI's performance was compared against expert‌ radiologists. The results suggest‌ that the algorithm can‌‌ provide a reliable, automated "second opinion," helping to‌ standardize the diagnosis of‌ chronic inflammatory diseases which‌‌ are often subjective and difficult to read.

8.31‌ Patritumab Deruxtecan in Breast‌ Cancer

Participants:Hugues Talbot‌‌ , Loïc Le Bescond (Collaboration: F. André, Gustave-Roussy)‌

This article 25 reports‌ on a Phase 2‌‌ clinical trial of a new antibody-drug conjugate (Patritumab‌ Deruxtecan) for patients with‌ specific types of advanced‌‌ breast cancer. While largely a medical study, unsupervised,‌ vision-based whoole-slide imaging techniques‌ were used to track‌‌ tumor response across patient groups, and more importantly,‌ at the slide level‌ to precisely indicate where‌‌ and on which cell groups the drug was‌ effective. This allowed the‌ article authors to propose‌‌ a biological pathway for the drug to operate.‌ Similar techniques are being‌ currently used to investigate‌‌ drugs in the same class. The trial demonstrates‌ promising efficacy in patients‌ whose cancer has become‌‌ resistant to standard therapies.‌

8.32 Evaluating generative models

Participants:Nicolas Salvy and‌ Hugues Talbot (Collaboration: B. Thirion, Inria Saclay)

A‌ recent position paper at ICML argued that all‌ quality metrics for image generation (using e.g. autoencoders,‌ GANs or diffusion models) are wrong. In response,‌ this paper 52 proposes two new metrics—Clipped Density‌ and Coverage—to better evaluate the quality of generative‌ AI models. Traditional metrics often fail to accurately‌ capture whether a model is simply "memorizing" data‌ or if it is producing truly diverse samples.‌ These new metrics provide a more nuanced look‌ at how well a model covers the data‌ distribution and the fidelity of the images it‌ creates. The position paper proposes a series of‌ benchmarks that a good-quality metric should pass. Our‌ proposal outperforms all existing methods on those benchmarks.‌

8.33 Estimating bone and muscle from low-dose X-Ray‌

Participants:Hugues Talbot (Collaboration: NAIST)

This research 45‌, the EOS low-dose X-ray system to estimate‌ bone mineral density (BMD) and muscle mass simultaneously.‌ By applying deep learning to these low-radiation 2D‌ images, we can predict body composition metrics that‌ typically require a more expensive and higher-radiation dual-energy‌ (DXA) scan, which derives bone density from the‌ varying absorption at the two beam energies. This‌ has significant implications for screening osteoporosis and other‌ bone, joint or even soft-tissue diseases (e.g. sarcopenia)‌ in aging populations.

9 Bilateral contracts and grants‌ with industry

9.1 Bilateral contracts with industry

PhD‌ Contract with Socotec (CIFRE Inria)
- Project title: Generation‌ of BIM models from two-dimensional architectural plans
- Duration:‌ 2025-2028
- Leader: Emilie Chouzenoux

PhD Contract with Entalpic‌ (CIFRE CentraleSupélec)
- Project title: Machine learning for materials‌ discovery
- Duration: 2025-2028
- Leader: Fragkiskos Malliaros and Hugues‌ Talbot

PhD Contract with SafranTech (CIFRE Inria)
- Project‌ title: Separation of vibratory and acoustic sources using‌ generative models for the aeronautical industry
- Duration: 2025-2028‌
- Leader: Emilie Chouzenoux

PhD Contract with Nokia Bell‌ Labs (CIFRE CentraleSupélec)
- Project title: Graph Neural Networks‌ for Causal Inference for Wireless Network Management
- Duration:‌ 2024-2027
- Leader: Fragkiskos Malliaros

PhD Contract with Heartflow‌ Inc (Inria)
- Project title: Estimating heart perfusion based‌ on physics-aware machine-learning methods
- Duration: 2022-2025
- Leaders: Hugues‌ Talbot and I. Vignon-Clementel (Inria REO).

PhD Contract‌ with SAFRAN (CIFRE CentraleSupélec)
- Project title: Safran fibres‌ composites
- Duration: 2023-2026
- Leaders: Hugues Talbot

PhD Contract‌ with GE Healthcare (CIFRE CentraleSupélec)
- Project title: Motion‌ correction in 3D X-ray interventional imaging of cerebrovascular‌ accidents
- Duration: 2025-2028
- Leaders: Jean-Christophe Pesquet and Nora‌ Ouzir

PhD Contract with Framatome (CIFRE CentraleSupélec)
- Project‌ title: Uncertainty quantification in neural networks for critical‌ applications
- Duration: 2025-2028
- Leaders: Jean-Christophe Pesquet

Contract with‌ Schneider Electric (CentraleSupélec and Telecom Sud Paris)
- Project‌ title: Detection of pump cavitation using artificial intelligence‌ algorithms
- Duration: 2024-2025
- Leaders: Jean-Christophe Pesquet and Marc‌ Castella

10 Partnerships and cooperations

10.1 International initiatives‌

10.1.1 STIC/MATH/CLIMAT AmSud projects

CGLFRVE

Title:
Context-guided future‌ liver remnant volume estimation using artificial intelligence models‌
Program:
STIC-AmSud
Duration:
January 1, 2024 – December‌ 31, 2025
Local supervisor:
Maria Vakalopoulou
Partners:
- Chang (Chili)
- Ferrante (Argentine)
- Universidad‌ de los Andes
Inria‌ contact:
Maria Vakalopoulou
Summary:‌‌
Automatic Liver-Segmentation is an essential task in the‌ medical context. Current AI-based‌ models are focused on‌‌ liver and tumor segmentation, that is not enough‌ for surgical planning, especially‌ for liver metastases. An‌‌ automatic liver and tumor segmentation method can greatly‌ relieve physicians of the‌ heavy workload of examining‌‌ CT images. However, for surgery, a more challenging‌ task is required. In‌ this context, it is‌‌ critical to estimate accurately the remnant liver volume‌ after resection; for instance,‌ in patients with liver‌‌ metastases. Estimating the future liver remnant is a‌ challenging task because the‌ type of surgery to‌‌ be performed depends on each patient’s clinical setting,‌ the center’s experience, number‌ and location of liver‌‌ lesions, among others. This means that future liver‌ remnant segmentation depends on‌ the patient’s clinical context.‌‌ Therefore, the goal of this project is to‌ design, implement and evaluate‌ fine-grained liver segmentation guided‌‌ by the context that allows us to precisely‌ estimate remnant liver volume.‌ Our work is guided‌‌ by five objectives: (1) evaluate SOTA liver segmentation‌ models, including the recent‌ published architecture HybridGNet; (2)‌‌ design and evaluate models for fine-grained liver segmentation‌ models taking into account‌ models like SAM and‌‌ HybridGNet; (3) estimate remnant liver volume using the‌ fine-grained liver segmentation model;‌ (4) Integrate contextual information‌‌ by prompts for liver segmentation. Finally, we present‌ results on public and‌ private datasets. For the‌‌ private case, we collaborate with a local health‌ center, which provides us‌ access to data. To‌‌ accomplish the proposed objectives, we have formed a‌ multidisciplinary team, including physicians‌ with specialization in radiology‌‌ and experts on computer vision applied to medical‌ images.

10.2 International research‌ visitors

10.2.1 Visits of‌‌ international scientists

International visits to the team

Vlad‌ Vasilescu

Status
PhD
Institution‌ of origin:
University Polytechnique‌‌ of Bucharest
Country:
Romania
Dates:
7th Feb. to‌ 7th April 2025
Context‌ of the visit:
Collaboration‌‌ with Jean-Christophe Pesquet
Mobility program/type of mobility:
Research‌ stay funded by ERASMUS‌

Subhajit Saha

Status
PhD‌‌
Institution of origin:
TCG Crest
Country:
India
Dates:‌
6th Feb. to 4th‌ April 2025
Context of‌‌ the visit:
Collaboration with Emilie Chouzenoux
Mobility program/type‌ of mobility:
Research stay‌ funded by ERC MAJORIS‌‌

Jose Manuel Saavedra

Status
Professor
Institution of origin:‌
Universidad de los Andes‌
Country:
Chile
Dates:
15th‌‌ Jan. 2025 to 31st Jan. 2025
Context of‌ the visit:
Collaboration with‌ Maria Vakalopoulou .
Mobility‌‌ program/type of mobility:
Research stay through Inria international‌ mobility funding under STIC-AmSud‌ project CGLFRVE.

Violeta Chang‌‌

Status
Professor
Institution of origin:
Universidad de los‌ Andes
Country:
Chile
Dates:‌
15th Jan. 2025 to‌‌ 31st Jan. 2025
Context of the visit:
Collaboration‌ with Maria Vakalopoulou .‌
Mobility program/type of mobility:‌‌
Research stay through Inria international mobility funding under‌ STIC-AmSud project CGLFRVE.

10.2.2‌ Visits to international teams‌‌

Research stays abroad

Jean-Christophe Pesquet

Visited institution:
Danish‌ Technical University
Country:
Denmark‌
Dates:
15th Nov. 2025‌‌ to 15th Dec. 2025‌
Context of the visit:
Collaboration with CUQI (Computational‌ Uncertainty Quantification for Inverse problems) research group
Mobility‌ program/type of mobility:
Research stay funded by Otto‌ Monsted visiting professorship

David Restrepo

Visited institution:
CONICET–University‌ of Buenos Aires
Country:
Argentina
Dates:
14th Dec.‌ 2025 to 3rd Dec. 2025
Context of the‌ visit:
Research visit to collaborate with Enzo Ferrante,‌ faculty researcher at CONICET–University of Buenos Aires, on‌ a project related to biases in vision–language models.‌
Mobility program/type of mobility:
Research stay supported through‌ Inria international mobility funding under the STIC-AmSud program‌ CGLFRVE.

10.3 European initiatives

10.3.1 Horizon Europe

EDUCADO‌

EDUCADO project on Cordis.europa.eu

Title:
Exploring the Deep‌ Universe by Computational Analysis of Data from Observations‌
Duration:
1 January 2024 to 31 December 2027‌
Partners:
- Université Gustave-Eiffel (FR)
- Université de Groningen (NL)‌
- Université de Ghent (BEL)
- University of Napoli (ITA)‌
- University of Barcelona (ESP)
- Instituto de Astrofisica de‌ Canarias (ESP, coordinator)
Coordinator:
Instituto de Astrofisica de‌ Canarias (ESP, coordinator)
Role of OPIS:
Hugues Talbot‌ (participant, through UGE partner)
Summary:
In recent years,‌ there has been a significant increase in space‌ research and observations. Better equipment has been developed‌ and introduced, and more resources are being used.‌ However, despite these advancements, many faint galaxies can‌ still go unnoticed. Therefore, innovative developments in information‌ technology are needed. The MSCA-funded EDUCADO project aims‌ to foster extensive collaboration between the fields of‌ astronomy and computer science. This collaboration will leverage‌ the expertise of specialists from both sectors to‌ detect previously unseen numbers of faint observable galaxies,‌ analyse the frequency, types and characteristics of dwarf‌ galaxies, and compare them to the Milky Way.‌ The project will also provide training for 10‌ doctoral candidates to develop high-quality methods in these‌ fields.

10.3.2 H2020 projects

MAJORIS

MAJORIS project on‌ cordis.europa.eu

Title:
Majoration-Minimization algorithms for Image Processing
Duration:‌
From January 1, 2020 to June 30, 2025‌
Partners:
- Inria Saclay (FR)
Coordinator:
Emilie Chouzenoux
Summary:‌

Mathematical optimization is the key to solving many‌ problems in science, based on the observation that‌ physical systems obey a general principle of least‌ action. While some problems can be solved analytically,‌ many more can only be solved via numerical‌ algorithms. Research in this domain has proved essential‌ over many years. In addition, science in general‌ is changing. Increasingly, in biology, medicine, astronomy, chemistry,‌ physics, large amounts of data are collected by‌ constantly improving signal and image acquisition devices, that‌ must be analyzed by sophisticated optimization tools. In‌ this proposal, we consider handling optimization problems with‌ large datasets. This means minimizing a cost function‌ with a complex structure and many variables. The‌ computational load for solving these problems is too‌ great for even state-of-the-art algorithms. Thus, only relatively‌ rudimentary data processing techniques are employed, reducing the‌ quality of the results and limiting the outcomes‌ that can be achieved via these novel instruments.‌ New algorithms must be designed with computational scalability,‌ robustness and versatility in mind.

In this context, Majorization-Minimization (MM) approaches have‌ a crucial role to‌ play. They consist of‌‌ a class of efficient and effective optimization algorithms‌ that benefit from solid‌ theoretical foundations. The MAJORIS‌‌ project aims at proposing a breakthrough in MM‌ algorithms, so that they‌ remain efficient when dealing‌‌ with big data. I propose to tackle several‌ challenging questions concerning algorithm‌ design. These include acceleration‌‌ strategies, convergence analysis with complex costs and inexact‌ schemes. I will also‌ tackle practical, massively parallel‌‌ and distributed architecture implementations. Three specific applications are‌ targeted: super-resolution in multiphoton‌ microscopy in biology; on-the-fly‌‌ reconstruction for 3D breast tomosynthesis in medical imaging;‌ and mass spectrometry data‌ processing in chemistry.

10.4‌‌ National initiatives

10.4.1 ANR

Program: ANR (appel générique‌ CE45)
- Project acronym: AAIMME‌
- Project title: Apprentissage Automatique‌‌ pour l'Imagerie Moléculaire et la MEdecine du futur‌
- Duration: 2025-2028
- Coordinator: Geoffrey‌ Daniel (DM2S, IRFU), local‌‌ coordinator: Emilie Chouzenoux , participants: Jean-Christophe Pesquet ,‌ Ludovic Trautmann

Program: ANR‌ (appel générique)
- Project acronym:‌‌ Delta
- From DEep Learning to clinical Tissue Anisotropy:‌ proof of concept infarct‌ lesion characterisation with 3D‌‌ ultrasound
- Duration: 2025-2029
- Coordinator: François Varray (Insa Lyon),‌ local coordinator: Nora Ouzir‌

Program: FHU
- Project acronym:‌‌ Mosaic 2
- Multiscale Optimised Strategy for Artificial intelligence-based‌ Imaging biomarkers in digestive‌ Cancer
- Duration: 2025-2029
- Coordinator:‌‌ Valerie Paradis (Hôpital Beaujon, APHP), local coordinator: Nora‌ Ouzir

Program: FHU
- Project‌ acronym: Transvir
- Innovative strategies‌‌ to treat chronic viral infections after hematopoietic stem‌ cell and solid organ‌ transplantation
- Duration: 2025-2029
- Coordinator:‌‌ Le Goff Jérôme (Hôpital Saint-Louis, APHP), local coordinator:‌ Nora Ouzir

Program: ANR‌ RHU
- Project acronym: Innov4-ePiK‌‌
- Project title: Innovative diagnostic and therapeutic approaches in‌ epileptic and developmental encephalopathies‌ linked to potassium channel‌‌ genes using the 4P framework for medicine.
- Duration:‌ 2024-2028
- Coordinator: Rima Nabbout,‌ Université Paris Cité /‌‌ AP-HP, local: Fragkiskos Malliaros , participant: Nora Ouzir‌

Program: ANR JCJC
- Project‌ acronym: Hagnodice
- Project title:‌‌ Holistic explainable artificial intelligence schemes for lung cancer‌ prognosis
- Duration: 2022-2026
- Coordinator:‌ Maria Vakalopoulou

Program: ANR‌‌ JCJC
- Project acronym: GraphIA
- Project title: Scalable and‌ robust representation learning on‌ graphs
- Duration: 2021-2025
- Coordinator:‌‌ Fragkiskos Malliaros

Program: ANR JCJC
- Project acronym: SIMPLES‌
- Project title: New Methods‌ for Nonsmooth Stochastic Bilevel‌‌ Optimization
- Duration: 2025-2029
- Coordinator: Antonio Silveti-Falls

10.4.2 PIQ‌

Program: PIQ Inria
- Project‌ acronym: KhiMalMa
- Project title:‌‌ Towards non-invasive screening for muscle diseases using novel‌ non-linear optics combined with‌ advanced data processing methods‌‌
- Duration: 2025-2027
- Coordinator: Claire Lefort (XLIM, CNRS), local:‌ Emilie Chouzenoux , participants:‌ Jean-Christophe Pesquet , Samy‌‌ Ferrat

10.4.3 Others

PRISM

Participants:Hugues Talbot ,‌ Loïc Le Bescond ,‌ Maria Vakalopoulou (Collaboration: S.‌‌ Christodoulidis, P.-H. Cournède, MICS, CentraleSupélec ; F. André,‌ Gustave-Roussy)

The PRISM program‌ at Gustave-Roussy, which is‌‌ a major research program on precision medicine, has‌ been funded by the‌ ANR since 2018 by‌‌ a grant of €5 millions. This was considered‌ a major achievement for‌ the team, as it‌‌ has allowed us to continue our research on‌ the use of AI‌ for precision medicine.

It‌‌ has now received Institut‌ Hospitalo-Universitaire (IHU) label. The vision of the project‌ is transformative in its approach for cancer treatment.‌ It aims to better understand the biology of‌ each patient’s cancer and to identify, from diagnosis,‌ those with the most aggressive tumours in order‌ to offer them the most appropriate treatment. This‌ IHU label is part of the perspective of‌ making Gustave Roussy the largest campus in Europe‌ dedicated to cancer.

The PRISM program has become‌ one of the 5 IHU endowed with 30‌ to 40 million euros announced by the Government‌ as part of the 3rd call for projects‌ of the France 2030 plan. The objective of‌ the IHUs is to strengthen French medical research‌ capacity by developing world-class research (clinical and translational)‌ skills involving university, health establishment, research organizations and‌ companies.

PRISM is the result of several years‌ of research conducted by the teams of Gustave‌ Roussy in partnership with CentraleSupélec, Université Paris-Saclay, Inserm‌ and Unicancer.

DataIA Cluster

Project title: Pre-trained Models‌ for Data Frugality in Health Multimodal Few-shot Learning‌ for Medical Imaging
Duration: 2025-2029
Coordinators: Nora Ouzir‌ , Edouard Duchesnay (CEA) and Florent Bouchard (CNRS,‌ L2S)
Organization: DataIA Cluster, Chaire Synergie

FMJH

Project‌ title: Fast Optimization and Computational Understanding of Systems‌
Duration: 01/09/2023 - 31/08/2024
Coordinator: Antonio Silveti-Falls
Organization:‌ Fondation Mathématique Jacques Hadamard

11 Dissemination

11.1 Promoting‌ scientific activities

11.1.1 Scientific events: organisation

General chair,‌ scientific chair

Maria Vakalopoulou has served as a‌ Workshop Chair for ICCV 2025 in Honolulu, Hawai'i.‌

Member of the organizing committees

Emilie Chouzenoux :‌ Co-organizer of the recurrent workshop Mathematical Fundations of‌ AI (DATAIA and SCAI), 25 Mar. 2025, 10‌ Dec. 2025, Paris.
Mounir Kaaniche : Publications chair‌ at the European Workshop on Visual Information Processing‌ (EUVIP), Valleta, Malta, Oct. 2025.
Fragkiskos Malliaros :‌ Co-organizer, Learning on Graphs (LoG) Paris meet up,‌ 2025.
Fragkiskos Malliaros : Senior Program Committee and‌ Area Chair at Neural Information Processing Systems (NeurIPS),‌ ACM SIGKDD Conference on Knowledge Discovery and Data‌ Mining (KDD), and European Conference on Machine Learning‌ (ECML PKDD).
Hugues Talbot : Member of the‌ steering committee for the DGMM 2025 conference in‌ Groningen (NL).

11.1.2 Scientific events: selection

Chair of‌ conference program committees

Antonio Silveti-Falls : Area chair‌ of ICML and NeurIPS 2025.
Maria Vakalopoulou :‌ Area Chair at Computer Vision and Pattern Recognition‌ (CVPR) 2025, International Computer Vision and Pattern Recognition‌ (ICCV) 2025.
Hugues Talbot : Area Chair for‌ CVPR 2026.

Member of the conference program committees‌

Hugues Talbot : Senior program committee member for‌ AAAI 2026.

Reviewer

The members of the team‌ reviewed numerous papers for several international conferences, such‌ as for the annual conferences on Computer Vision‌ and Pattern Recognition (CVPR), Medical Image Computing and‌ Computer Assisted Intervention (MICCAI), Neural Information Processing Systems‌ (NeurIPS), IEEE InternationalWorkshop onMachine Learning for Signal Processing‌ (MLSP), International Conference on Learning Representations (ICLR), IEEE‌ International Conference and Acoustics Speech and Signal Processing‌ (ICASSP), IEEE International Conference on Image Processing (ICIP), IEEE Statistical Signal Processing‌ workshop (SSP), European Signal‌ Processing Conference (EUSIPCO), AAAI‌‌ Conference on Artificial Intelligence (AAAI), The Web Conference‌ (WWW), Annual Conference of‌ the North American Chapter‌‌ of the Association for Computational Linguistics (NAACL), International‌ Conference on Web and‌ SocialMedia (ICWSM), International Conference‌‌ onMachine Learning (ICML), Conference on Neural Information Processing‌ Systems (NeurIPS), International Conference‌ on Complex Networks and‌‌ Their Applications (Complex Networks), International Workshop on Graph-Based‌ Natural Language Processing (TextGraphs),‌ Artificial Intelligence and Statistics‌‌ Conference (AIStat), British Machine Vision Conference, Montreal AI‌ Symposium, ACM SIGKDD Conference‌ on Knowledge Discovery and‌‌ Data Mining (KDD), IEEE/ACM International Conference on Advances‌ in Social Networks Analysis‌ and Mining (ASONAM), and‌‌ Learning on Graphs Conference (LoG).

11.1.3 Journal

Member‌ of the editorial boards‌

Emilie Chouzenoux : Associate‌‌ Editor of the SIAM Journal on Imaging Sciences.‌
Emilie Chouzenoux : Associate‌ Editor of the SIAM‌‌ Journal on Mathematics of Data Sciences.
Fragkiskos Malliaros‌ : Associate Editor, Big‌ Data Research, Elsevier.
Fragkiskos‌‌ Malliaros : Guest Editor, Applied Network Science, Springer.‌
Fragkiskos Malliaros : Guest‌ Editor, Data Mining and‌‌ Knowledge Discovery, Springer.
Nora Ouzir : Associate Editor‌ of Elsevier Signal Processing.‌
Jean-Christophe Pesquet : Associate‌‌ Editor of the SIAM Journal on Imaging Sciences.‌
Hugues Talbot : Review‌ Editor for the journal‌‌ Frontiers in Computer Science.
Maria Vakalopoulou : Editor‌ at Computer Vision and‌ Image Understanding (CVIU) Journal.‌‌
Maria Vakalopoulou : Associate Editor at Medical Image‌ Analysis Journal.

Reviewer -‌ reviewing activities

The members‌‌ of the team reviewed numerous papers for several‌ international journals, such as‌ Inverse Problems, IEEE Transactions‌‌ on Signal Processing, IEEE Transactions on Computational Imaging,‌ IEEE Signal Processing Letters,‌ Signal Processing (Elsevier), International‌‌ Journal of Data Science and Analytics, Digital Signal‌ Processing, SIAM Journal of‌ Optimization, International Journal of‌‌ Computer Vision, Signal Processing: Image Communication, Computational Optimization‌ and Application.

11.1.4 Invited‌ talks

Emilie Chouzenoux :‌‌ Plenary speaker, SIAM Annual Meeting, Montreal, Canada, July‌ 2025
Emilie Chouzenoux :‌ Invited speaker, Séminaire Imagerie‌‌ en santé, méthodologies et problèmes ouverts, GS Santé‌ publique, Villejuif, Nov. 2025‌
Emilie Chouzenoux : Invited‌‌ speaker, Electrical and Electronic Engineering Department, Imperial College‌ London, UK, Oct. 2025‌
Emilie Chouzenoux : Invited‌‌ speaker, Numerical Analysis Seminar, Univ. Oxford, UK, Oct.‌ 2025
Emilie Chouzenoux :‌ Invited speaker, GDR IASIS‌‌ workshop on Unrolling and un/self/*/supervised learning for inverse‌ problems, Paris, May 2025‌
Emilie Chouzenoux : Invited‌‌ speaker, Journées Franco-Chiliennes Optimisation, Rouen, May 2025
Fragkiskos‌ Malliaros : Invited speaker,‌ Bellairs Workshop on Machine‌‌ Learning and Statistical Signal Processing, McGill University, Barbados,‌ 2025.
Fragkiskos Malliaros :‌ Kenote speaker, International Workshop‌‌ on Graph-Based Representations in Pattern Recognitions, Caen, France,‌ 2025.
Fragkiskos Malliaros :‌ Invited speaker, Machine Learning‌‌ on Graphs Working Group, Normandie Université, France, 2025.‌ (Online).
Jean-Christophe Pesquet :‌ Invited speaker at BASP‌‌ Frontiers Workshop, Villars-sur-Ollon, Switzerland, Jan. 2025.
Jean-Christophe Pesquet‌ : Tutorial speaker at‌ 33rd European Signal Processing‌‌ Conference (EUSIPCO 2025), Isola delle Femmine, Sicily, Italy,‌ Sep. 2025.
Hugues Talbot‌ : Invited speaker, SPRING‌‌ Saclay, May 21st 2025.‌
Hugues Talbot : Keynote speaker, inauguration of MesoSPIM‌ Paris-Saclay, June 6th 2025.
Hugues Talbot : Invited‌ speaker, Instituto de phisica de Canarias, Las Palmas,‌ June 26 2025.
Maria Vakalopoulou : Invited speaker,‌ AMLD, EPFL, Feb. 2025.
Maria Vakalopoulou : Invited‌ speaker, 57th ESPGHAN Annual Meeting, Helsinski, Finland, May‌ 2025.
Maria Vakalopoulou : Invited speaker, AI4EO symposium,‌ Rennes, France, Sep. 2025.
Maria Vakalopoulou : Invited‌ speaker, Machine Intelligence for iNverse imaging, Observation Analysis‌ and Sensing (MINOAS) Workshop, Heraklion, Greece, Sep. 2025.‌
Maria Vakalopoulou : Invited speaker, 1st International Workshop‌ on Biomedical Image and Signal Computing for Unbiasedness,‌ Interpretability and Trustworthiness, ICCV'2025, Honolulu, Hawai'i, Oct. 2025.‌

11.1.5 Leadership within the scientific community

Emilie Chouzenoux‌ : IEEE Senior Member, since September 2020.
Emilie‌ Chouzenoux : Associate member of the EURASIP Technical‌ Area Committee (TAC) on Signal and Data Analytics‌ for Machine Learning (SiG-DML).
Mounir Kaaniche : Elected‌ member of the EURASIP Technical Area Committee (TAC)‌ on Signal and Data Analytics for Machine Learning‌ (SiG-DML), since January 2025.
Mounir Kaaniche : IEEE‌ Senior Member, since September 2020.
Nora Ouzir :‌ Elected member of the EURASIP Technical Area Committee‌ (TAC) on Biomedical Image and Signals Analytics.
Jean-Christophe‌ Pesquet : Senior honorary member of the Institut‌ Universitaire de France and Fellow of IEEE and‌ EURASIP.

11.1.6 Scientific expertise

The members of the‌ team participated to numerous PhD Thesis Committees, PhD‌ “comité de suivi individuel”, HdR Committees, recruiting Committees,‌ and served as Grant Reviewers.

Emilie Chouzenoux :‌ member of the scientific committee of PIQ program‌ of Inria
Emilie Chouzenoux : member of the‌ executive committee of DataIA institute of University Paris‌ Saclay (until aug. 2025)
Emilie Chouzenoux : member‌ of the scientific committee of the programme Mathématiques‌ du Calcul Scientifique et de l'Ingénierie of the‌ Labex Mathématiques Hadamard (Maths CSI LMH)
Nora Ouzir‌ : elected member of CCUPS, Consultative Commission of‌ Paris-Saclay University
Nora Ouzir : member of the‌ scientific committee of the Programme Doctoral SCAI, Cluster‌ IA de Sorbonne Université
Maria Vakalopoulou : Monitor-expert‌ for EU research grants.
Alix Chazottes : Jury‌ member of the ANR Compétences & Métiers d'Avenir‌, plan de relance France2030.

11.1.7 Research administration‌

Jean-Christophe Pesquet is the head of the laboratoire‌ Centre pour la Vision Numérique (CVN), CentraleSupélec

11.1.8‌ Teaching administration

Emilie Chouzenoux : attached professor in‌ AI in CentraleSupélec, since sep. 2022.
Nora Ouzir‌ : Co-coordinator of the Bachelor in Artificial Intelligence,‌ Data and Management Sciences, AIDAMS, with ESSEC Business‌ School and CentraleSupélec.
Nora Ouzir : Co-coordinator of‌ the Thematic Sequence 2 (ST2) Viral Propagation, 1st‌ year of CentraleSupélec.
Nora Ouzir : Head for‌ the Machine Learning Elective course in 2nd year‌ of CentraleSupélec (210 students).
Fragkiskos Malliaros : co-director‌ of the Master in Data Sciences and Business‌ Analytics (DSBA).
Fragkiskos Malliaros : head of the‌ Data and Information Sciences (SDI) specialization at CentraleSupélec.‌
Fragkiskos Malliaros : academic responsible of CentraleSupélec's Summer‌ School on AI.
Jean-Christophe Pesquet : local head for the Optimization M.Sc.‌ at CentraleSupélec.
Jean-Christophe Pesquet‌ : head for the‌‌ main Optimization course in 2nd year of CentraleSupélec‌ (600 students).
Hugues Talbot‌ : head of the‌‌ Mastère Specialisé in "Trustworthy AI". It was in‌ development since 2022 with‌ IRT SystemX and CentraleSupelec‌‌ Exed. This Master is designed for professionals who‌ want to retrain in‌ AI and focuses on‌‌ the explainability and trustworthiness of models. It represents‌ 400h of teaching and‌ opened its first batch‌‌ this year.
Hugues Talbot : local head for‌ CentraleSupélec the Mathematiques, Vision‌ Apprentissage (MVA) Master of‌‌ Ecole Normale Supérieure Paris-Saclay

11.2 Teaching - Supervision‌ - Juries - Educational‌ and pedagogical outreach

11.2.1‌‌ Teaching

Several permanent members of OPIS were involved‌ as lecturer (lec.) or‌ lab instructors (lab), in‌‌ the following courses.

Master: Emilie Chouzenoux . Foundations‌ of Distributed and Large‌ Scale Computing, 26h (lec.),‌‌ 3rd year CentraleSupélec and M.Sc. MVA Paris Saclay.‌
Master: Emilie Chouzenoux .‌ Advanced Machine Learning, 18h‌‌ (lec.), 3rd year CentraleSupélec.
Bachelor: Emilie Chouzenoux .‌ Optimization, 24h (lec.), 1st‌ year Bachelor AIDAMS, Univ.‌‌ Paris Saclay.
Master/PhD: Emilie Chouzenoux . Introduction to‌ theory and numerics of‌ large scale optimization, 12h‌‌ (lec. + labs), CIMPA School on Optimal Transport,‌ PDEs and Optimization, Essaouira,‌ Marocco, May 2025
Master:‌‌ Nora Ouzir . Machine Learning, 30h (lec.), 2nd‌ year course, CentraleSupélec, EN‌
Master: Nora Ouzir .‌‌ Advanced Machine Learning, 24h (lec.), Msc DSBA, CentraleSupélec/ESSEC.‌
Master: Nora Ouzir .‌ Optimization, 15h (lab.), 2nd‌‌ year course, CentraleSupélec.
Bachelor: Nora Ouzir . Signal‌ Procesing, 10h (lab.), 1st‌ year course, CentraleSupélec.
Master:‌‌ Nora Ouzir . Medical Image Processing 3h (lec.),‌ AI and Global Health‌ program of the European‌‌ University Alliance for Global Health (EUGLOH).
Master: Fragkiskos‌ Malliaros . Machine Learning‌ in Network Science, 27h‌‌ (lec.), Master in Data Sciences and Business Analytics,‌ CentraleSupélec and ESSEC Business‌ School, M.Sc. in Artificial‌‌ Intelligence, CentraleSupélec, 3rd year data science mention, CentraleSupélec.‌
Master: Fragkiskos Malliaros .‌ Foundations of Machine Learning,‌‌ 48h (lec.), Master in Data Sciences and Business‌ Analytics, CentraleSupélec and ESSEC‌ Business School.
Master: Jean-Christophe‌‌ Pesquet . Introductory course on Optimization, 33h (lec.),‌ 2nd year CentraleSupélec.
Master:‌ Jean-Christophe Pesquet . Introduction‌‌ to Optimization, 6h (lec.), M.Sc. MVA Paris Saclay.‌
Master: Jean-Christophe Pesquet .‌ Advanced course on Optimization,‌‌ 10h (lec.), M.Sc. in Signal Processing and Automatic‌ Control, Univ. Paris-Saclay.
Master:‌ Jean-Christophe Pesquet . Convex‌‌ Optimization Algorithms, 15h (lec.), M.Sc. in Optimization, Univ.‌ Paris-Saclay.
Master: Antonio Silveti-Falls‌ . Convergence, Integration, and‌‌ Probability, 18h (lec.) + 18h (lab), 1st year‌ course, CentraleSupélec.
Master: Antonio‌ Silveti-Falls . Optimization for‌‌ Computer Vision, 21h (lec. + lab), 3rd year‌ course, CentraleSupélec.
Master: Antonio‌ Silveti-Falls . Partial Differential‌‌ Equations, 13.5h (lab), 1st year course, CentraleSupélec.
Master:‌ Antonio Silveti-Falls . Optimization,‌ 12.5h (lab), 2nd year‌‌ course, CentraleSupélec.
Bachelor: Antonio Silveti-Falls . Analysis I,‌ 18h (lab), First year‌ Bachelors of Global Engineering‌‌ students, Univ. Paris Saclay.
Bachelor: Antonio Silveti-Falls .‌ Analysis II, 18h (lab),‌ First year Bachelors of‌‌ Global Engineering students, Univ.‌ Paris Saclay.
Bachelor: Antonio Silveti-Falls . Analysis III,‌ 18h (lab), First year Bachelors of Global Engineering‌ students, Univ. Paris Saclay.
Bachelor: Antonio Silveti-Falls .‌ Probability, 18h (lab), First year Bachelors of Global‌ Engineering students, Univ. Paris Saclay.
Bachelor: Antonio Silveti-Falls‌ . Topology and Functional Analysis, 18h (lab), Second‌ year Bachelors of Global Engineering students, Univ. Paris‌ Saclay.
Master: Hugues Talbot . Convergence Integration Probabilité,‌ 18h (lab), 1st year course, CentraleSupélec.
Master: Hugues‌ Talbot . High-performance computing, 12h (lab), 2nd year‌ CentraleSupélec.
Master: Hugues Talbot . Introduction à la‌ morphologie mathématique: 12h (lab), 3rd year CentraleSupélec.
Master:‌ Hugues Talbot . Modern mathematical morphology (20h lec.‌ + 8h lab), 3rd Year CentraleSupélec and MVA‌ (ENS Paris-Saclay).
Master: Hugues Talbot . Optimisation for‌ AI, (20h lec + 8h lab) M.Sc in‌ AI, CentraleSupélec.
Master: Hugues Talbot . Introduction to‌ Machine Learning, MS Management of Technogy, 30h (lec.).‌
Master: Maria Vakalopoulou . Introduction to Visual Computing,‌ CentraleSupélec, 25h (lec).
Master: Maria Vakalopoulou . Introduction‌ to Deep Learning, M.Sc. in Data Sciences and‌ Business Analytics, CentraleSupélec and ESSEC Business School, 24h‌ (lec).
Master: Maria Vakalopoulou . Introduction to Deep‌ Learning, M.Sc. in Artificial Intelligence, CentraleSupélec, 24h (lec).‌
Master: Maria Vakalopoulou . Deep Learning, M.Sc. in‌ Vision and Machine Learning, ENS Paris-Saclay, 25h (lec).‌
Master: Maria Vakalopoulou . Deep Learning in Medical‌ Imaging, M.Sc. in Vision and Machine Learning, ENS‌ Paris-Saclay, 25h (lec).
Master: Maria Vakalopoulou . Introduction‌ to Deep Learning, M.Sc. in Artificial Intelligence, CentraleSupélec,‌ FR, 24 h (lec).

Several students members of‌ OPIS have teaching assistant activities, in the following‌ cursus of the Univ. Paris Saclay campus:

Bachelor‌ AIDAMS, CentraleSupélec and ESSEC
ENSTA ParisTech
CentraleSupélec
M.Sc.‌ DSBA, CentraleSupélec and ESSEC
M.Sc. MVA, Univ. Paris‌ Saclay

11.2.2 PhD supervision

PhD (completed): Loïc Le‌ Bescond . Precision medicine, Histology and Deep learning,‌ 2021-2024, supervised by F. André (IGR) and Hugues‌ Talbot .
PhD (completed): Thomas Guilmeau . Algorithmes‌ stochastiques pour l'optimisation non convexe, 2021-2024, supervised by‌ Emilie Chouzenoux and V. Elvira (Univ. Edinburgh).
PhD‌ (in progress): Andrea Persici. Semantic Analysis of Deep-Sky‌ Images using Machine Learning and Structural Approaches, 2025-2027,‌ supervised by Benjamin Perret (UGE) et Hugues Talbot‌ .
PhD (in progress): Raaja El Hamdani .‌ Robust graph representation learning and applications in misinformation‌ detection, 2021-2024, supervised by Fragkiskos Malliaros and T.‌ Bonald (Télécom-Paris).
PhD (in progress): Clement Cosserat .‌ Algorithmes de majoration-minimisation pour le traitement du signal‌ statistique, 2022-2025, supervised by Emilie Chouzenoux and T.‌ Adali (Univ. Baltimore, USA).
PhD (in progress): Alix‌ Chazottes . Algorithmes d’optimisation dépliés pour la reconstruction‌ d’images à partir de données TEP dynamique, 2023-2026,‌ supervised by Emilie Chouzenoux and F. Sureau (CEA,‌ Biomaps).
PhD (in progress): Aymen Sardroui . Histopathological‌ image analysis, 2022-2025, supervised by Mounir Kaaniche and‌ Jean-Christophe Pesquet .
PhD (in progress): Arsene Amoya.‌ Neural networks-based stereo image retrieval, 2023-2026, supervised by‌ Mounir Kaaniche and A. Benazza-Benyahia (SUP'COM-Tunis).
PhD (in‌ progress): Nabil Mouadden. Deep Learning methods for lung applications, 2023-2026, supervised by‌ Maria Vakalopoulou and G.‌ Chassagnon (Hôpital Cochin).
PhD‌‌ (in progress): Jinqwei Zhang. Deep Learning methods on‌ Digital Pathology, 2020-2024, supervised‌ by Maria Vakalopoulou ,‌‌ and D. Samaras (Stony Brook University).
PhD (in‌ progress): Yassine Abbahaddou .‌ Topics in Geometric Deep‌‌ Learning, 2022-2025, supervised by Fragkiskos Malliaros , J.‌ Lutzeyer, and M. Vazirgiannis‌ (École Polytechnique).
PhD (in‌‌ progress): Vahan Martirosyan . Deep Graph Neural Networks‌ (GNNs) and Applications in‌ Biomedicine, 2023-2026, supervised by‌‌ Fragkiskos Malliaros , Hugues Talbot , and J.‌ Giraldo (Télécom Paris).
PhD‌ (in progress): Nicolas Salvy‌‌ . Génération d'images cérébrales fonctionnelles à grande échelle‌ poru améliorer la cartographie‌ cérébrale, 2023-2026, supervised by‌‌ Hugues Talbot , and Bertrand Thirion (Inria Saclay,‌ MIND).
PhD (in progress):‌ Hafsa El Herichi .‌‌ Extraction du tissage 3D des pièces composites de‌ grandes dimensions à partir‌ d’images tomographiques, 2023-2026, supervised‌‌ by Hugues Talbot , and Stéphane Roux (Laboratoire‌ de Mécanique de Paris-Saclay,‌ ENS Paris-Saclay).
PhD (in‌‌ progress): David Restrepo. Bias Analysis on vision and‌ language models 2024-2027, supervised‌ by Maria Vakalopoulou ,‌‌ Stergios Christodoulidis, Enzo Ferrante and Gilles Fay, MICS,‌ CentraleSupélec.
PhD (in progress):‌ Adam Ghalem. Graph Neural‌‌ Networks for Causal Inference for Wireless Network Management,‌ 2024-2027, supervised by Fragkiskos‌ Malliaros .
PhD (in‌‌ progress): Vuk Ignjatovic . Generative Models for Motion‌ Correction in Computed Tomography,‌ 2024-2027, supervised by Nora‌‌ Ouzir , Jean-Christophe Pesquet , and Cyril Riddell‌ (GE Healthcare).
PhD (in‌ progress): Raoul Sallé de‌‌ Chou, Machine-learning based Prediction of heart perfusion maps,‌ 2021-2025, supervised by Laurent‌ Najman (UGE), Hugues Talbot‌‌ , Irene Vignon-Clémentel (Inria SimbotX)
PhD (in progress):‌ Francesco Songia. Reduced order‌ modelling of hemodynamics for‌‌ liver surgery procedure, 2025-2028, supervised by Nicolas Golse‌ (APHP), Irene Vignon-Clementel (Inria‌ SimbiotX), Hugues Talbot .‌‌
PhD (in progress): Imed Moussa . Vibratory and‌ acoustic source separation using‌ generative models for the‌‌ aerospace industry, 2025-2027, supervised by Emilie Chouzenoux and‌ Maxime Leiber (SafranTech).
PhD‌ (in progress): Paul Delage‌‌ . Generation of BIM models based on two-dimensional‌ architectural drawings, 2025-2027, supervised‌ by Emilie Chouzenoux and‌‌ Gottried Jacquet (SOCOTEC).
PhD (in progress): Ludovic Trautmann‌ . Uncertainty Quantification for‌ PET reconstructed images with‌‌ AI, supervised by Emilie Chouzenoux and Florent Sureau‌ (BioMaps).
PhD (in progress):‌ Shuai Mao . Robust‌‌ Few-shot Learning for Medical Imaging, 2025-2028, supervised by‌ Nora Ouzir and Jean-Christophe‌ Pesquet .
PhD (in‌‌ progress): Eve Delegue . Novel AI Methods for‌ Liver Cancer Histopathology Image‌ Analysis, 2025-2028, supervised by‌‌ Nora Ouzir , Jean-Christophe Pesquet , and Astrid‌ Laurent-Bellue (Bicêtre Hospital)

11.2.3‌ Intern/Engineers/Apprentices supervision

Maxence Adly‌‌ , Sep. 2024-Feb. 2025, supervised by Emilie Chouzenoux‌ (intern)
Idriss Benkirane ,‌ Sep. 2024-Feb. 2025, supervised‌‌ by Hugues Talbot (intern)
Benjamin Clene , Nov.‌ 2024-Jan. 2025, supervised by‌ Hugues Talbot (intern)
Bilal‌‌ Zidna , Mar. 2025-Aug. 2025, supervised by Emilie‌ Chouzenoux (intern)
Matthieu Merigot–Lombard‌ , May 2025 -‌‌ Sep. 2025, supervised by Emilie Chouzenoux (intern)
Bruno‌ Amorim De Araujo ,‌ April 2025 - Sep.‌‌ 2025, supervised by Emilie‌ Chouzenoux (intern)
Yassine Elammari , Jun. 2025-Dec. 2025,‌ supervised by Nora Ouzir (intern)
Eve Delegue ,‌ Apr. 2025-Aug. 2025, supervised by Nora Ouzir (intern)‌
Luis Evrard , Apr. 2025- Nov.2025, supervised by‌ Nora Ouzir (intern)
Alexandre Bertot , Sept. 2025-‌ Jan. 2026, supervised by Nora Ouzir (intern)
Ludovic‌ Trautmann , Dec. 2024 - Aug. 2025, supervised‌ by Emilie Chouzenoux (engineer)
Mohammad Mehdi Kalla ,‌ Sep. 2024-Aug. 2026, supervised by Emilie Chouzenoux (engineer‌ apprentice)
Samy Ferrat , Oct. 2025-Sep. 2027, supervised‌ by Emilie Chouzenoux (engineer)
Mohamed Salim Ben Omrane‌ , April - Sept. 2025, supervised by Mounir‌ Kaaniche and Jean-Christophe Pesquet (intern)

11.2.4 Juries

The‌ faculty members of the team serve regularly as‌ a jury Member to Final Engineering Internship and‌ the Research Innovation Project for students of CentraleSupélec,‌ and to Research Internship for students of Ms.C.‌ MVA, ENS Paris Saclay.

11.2.5 Educational and pedagogical‌ outreach

Alix Chazottes : Participation at outreach animation‌ events for high school students, through La Recherche‌ en Basket
Alix Chazottes : Presentation to ENS‌ Paris Saclay L3 students, through `Panorama recherche' vulgarization‌ event organized by Prof. L. Oudre.

11.3 Popularization‌

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

Emilie Chouzenoux : Participation to the dissemination paper‌ “Les maths à la rescousse de l’intelligence artificielle”‌ [link], Le Monde Cahier Sciences, 27 Oct.‌ 2025

11.3.2 Participation in Live events

Nora Ouzir‌ , Aymen Sardroui , Emilie Chouzenoux : Scientific‌ Days AI Action Summit 2025
Emilie Chouzenoux :‌ Premières Rencontres du Programme Inria Quadrant, June 2025‌

12 Scientific production

12.1 Major publications

1 article‌G.Guillaume Chassagnon, M.Maria Vakalopoulou,‌ E.Enzo Battistella, S.Stergios Christodoulidis,‌ T.-N.Trieu-Nghi Hoang-Thi, S.Severine Dangeard,‌ E.Eric Deutsch, F.Fabrice Andre,‌ E.Enora Guillo, N.Nara Halm,‌ S.Stefany El Hajj, F.Florian Bompard‌, S.Sophie Neveu, C.Chahinez Hani‌, I.Ines Saab, A.Aliénor Campredon‌, H.Hasmik Koulakian, S.Souhail Bennani‌, G.Gael Freche, M.Maxime Barat‌, A.Aurelien Lombard, L.Laure Fournier‌, H.Hippolyte Monnier, T.Téodor Grand‌, J.Jules Gregory, Y.Yann Nguyen‌, A.Antoine Khalil, E.Elyas Mahdjoub‌, P.-Y.Pierre-Yves Brillet, S.Stéphane Tran‌ Ba, V.Valérie Bousson, A.Ahmed‌ Mekki, R.-Y.Robert-Yves Carlier, M.-P.Marie-Pierre‌ Revel and N.Nikos Paragios. AI-driven quantification,‌ staging and outcome prediction of COVID-19 pneumonia.‌Medical Image Analysis67January 2021, 101860‌HAL DOI
2 articleP.Patrick Combettes and‌ J.-C.Jean-Christophe Pesquet. Fixed Point Strategies in‌ Data Science.IEEE Transactions on Signal Processing‌692021, 3878-3905HAL DOI
3 article‌N.Nathalie Lassau, S.Samy Ammari,‌ E.Emilie Chouzenoux, H.Hugo Gortais,‌ P.Paul Herent, M.Matthieu Devilder,‌ S.Samer Soliman, O.Olivier Meyrignac, M.-P.Marie-Pauline Talabard,‌ J.-P.Jean-Philippe Lamarque,‌ R.Remy Dubois,‌‌ N.Nicolas Loiseau, P.Paul Trichelair,‌ E.Etienne Bendjebbar,‌ G.Gabriel Garcia,‌‌ C.Corinne Balleyguier, M.Mansouria Merad,‌ A.Annabelle Stoclin,‌ S.Simon Jegou,‌‌ F.Franck Griscelli, N.Nicolas Tetelboum,‌ Y.Yingping Li,‌ S.Sagar Verma,‌‌ M.Matthieu Terris, T.Tasnim Dardouri,‌ K.Kavya Gupta,‌ A.Ana Neacsu,‌‌ F.Frank Chemouni, M.Meriem Sefta,‌ P.Paul Jehanno,‌ I.Imad Bousaid,‌‌ Y.Yannick Boursin, E.Emmanuel Planchet,‌ M.Mikael Azoulay,‌ J.Jocelyn Dachary,‌‌ F.Fabien Brulport, A.Adrian Gonzalez,‌ O.Olivier Dehaene,‌ J.-B.Jean-Baptiste Schiratti,‌‌ K.Kathryn Schutte, J.-C.Jean-Christophe Pesquet,‌ H.Hugues Talbot,‌ E.Elodie Pronier,‌‌ G.Gilles Wainrib, T.Thomas Clozel,‌ F.Fabrice Barlesi,‌ M.-F.Marie-France Bellin and‌‌ M.Michael Blum. Integrating deep learning CT-scan‌ model, biological and clinical‌ variables to predict severity‌‌ of COVID-19 patients.Nature Communications12634‌January 2021HAL DOI‌

12.2 Publications of the‌‌ year

International journals

4 articleS.Samy Ammari‌, A.Arnaud Quillent‌, V.Víctor Elvira‌‌, F.François Bidault, G.Gabriel Garcia‌, D.Dana Hartl‌, C.Corinne Balleyguier‌‌, N.Nathalie Lassau and É.Émilie Chouzenoux‌. Using Machine Learning‌ on MRI Radiomics to‌‌ Diagnose Parotid Tumours Before Comparing Performance with Radiologists:‌ A Pilot Study.‌Journal of Imaging Informatics‌‌ in Medicine383June 2025, 1496-1508‌HAL DOI back to‌ text
5 articleY.‌‌Younes Belkouchi, J.-C.Jean-Christophe Pesquet, A.‌Audrey Repetti and H.‌Hugues Talbot. Learning‌‌ truly monotone operators with applications to nonlinear inverse‌ problems: Learning truly monotone‌ operators.SIAM Journal‌‌ on Imaging Sciences1812025, 735-764‌HAL back to text‌
6 articleH.Hakim‌‌ Benkirane, M.Maria Vakalopoulou, D.David‌ Planchard, J.Julien‌ Adam, K.Ken‌‌ Olaussen, S.Stefan Michiels and P.-H.Paul-Henry‌ Cournède. Multimodal CustOmics:‌ A unified and interpretable‌‌ multi-task deep learning framework for multimodal integrative data‌ analysis in oncology.‌PLoS Computational Biology21‌‌6June 2025, e1013012HAL DOI
7‌ articleJ.Jhon Castro-Correa‌, M.Mohsen Badiey‌‌, J.Jhony Giraldo and F.Fragkiskos Malliaros‌. Semi-supervised graph learning‌ for underwater source localization‌‌ using ship-of-opportunity spectrograms.Journal of the Acoustical‌ Society of America158‌3September 2025,‌‌ 1836-1848HAL DOI
8 articleS.Sayantika Chatterjee‌, S.Stuti Jain‌, K.Kriti Kumar‌‌, E.Emilie Chouzenoux and A.Angshul Majumdar‌. Deep Probabilistic Matrix‌ Factorization on Graphs: Application‌‌ to Drug Repositioning in Antimicrobial Resistance.IEEE/ACM‌ Transactions on Computational Biology‌ and Bioinformatics2025.‌‌ In press. HAL DOIback to text back‌ to text
9 article‌E.Emilie Chouzenoux,‌‌ C.Cecile Della Valle‌ and J.-C.Jean-Christophe Pesquet. Stability Bounds for‌ the Unfolded Forward-Backward Algorithm.Journal of Mathematical‌ Imaging and Vision67482025HAL DOI‌back to text back to text
10 article‌C.Clément Cosserat, B.Ben Gabrielson,‌ E.Emilie Chouzenoux, J.-C.Jean-Christophe Pesquet and‌ T.Tülay Adali. An Effective Iterative Solution‌ for Independent Vector Analysis with Convergence Guarantees.‌IEEE Transactions on Signal Processing732025,‌ 4718 - 4733HALDOI back to text‌back to text
11 articleB.Benjamin Cox‌, E.Emilie Chouzenoux and V.Víctor Elvira‌. GraphGrad: Efficient Estimation of Sparse Polynomial Representations‌ for General State-Space Models.IEEE Transactions on‌ Signal Processing73March 2025, 1562-1576HAL‌DOI back to textback to text
12‌ articleT.Tassnim Dardouri, M.Mounir Kaaniche‌, A.Amel Benazza-Benyahia and G.Gabriel Dauphin‌. Multi-Task Convolution Neural Network-Based Lifting Scheme for‌ Image Compression.Pattern Recognition Letters2025HAL‌back to text
13 articleR.Roland Eid‌, A.Anthony Tarabay, P.Pierre Decazes‌, C.Clémence David, F.Fouad Kerbage‌, J.Jean Zeghondy, L.Leony Antoun‌, C.Cristina Smolenschi, A.Alina Fuerea‌, M.Marine Valery, V.Valerie Boige‌, M.Maximiliano Gelli, L.Lambros Tselikas‌, J.Jerome Durand-Labrunie, Y.Younes Belkouchi‌, L.Lawrance Littisha, S.Samy Ammari‌, M.Michel Ducreux, N.Nathalie Lassau‌ and A.Antoine Hollebecque. Predictive factors of‌ FOLFIRINOX chemotherapy toxicity in pancreatic adenocarcinoma patients.‌Future Oncology216February 2025, 691-697‌HAL DOI
14 articleV.Víctor Elvira,‌ E.Emilie Chouzenoux and O.O Deniz Akyildiz‌. A Proximal Newton Adaptive Importance Sampler.‌IEEE Signal Processing Letters322025, 1545-1549‌HAL back to text
15 articleJ.-B.Jean-Baptiste‌ Fest, A.Audrey Repetti and E.Emilie‌ Chouzenoux. A stochastic use of the Kurdyka-Lojasiewicz‌ property: Investigation of optimization algorithms behaviours in a‌ non-convex differentiable framework.Foundations of Data Science‌2025. In press. HAL back to text‌
16 articleT.Thomas Guilmeau, E.Emilie‌ Chouzenoux and V.Víctor Elvira. Regularized Rényi‌ divergence minimization through Bregman proximal gradient algorithms.‌Journal of Machine Learning Research261572025‌, 1−56HAL back to text
17 article‌M.Martinez Leonard, F.François Andrieu,‌ F.Frédéric Schmidt, H.Hugues Talbot and‌ B.Bentley Mark. Robust automatic crater detection‌ at all latitudes on Mars with Deep-learning.‌Planetary and Space Science260June 2025HAL‌DOI back to text
18 articleH.Huan‌ Li, X.Xinpeng Huang, Y.Yilei‌ Chen, C.Chao Yang, M.Mounir‌ Kaaniche, Q.Qiuwen Zhang and P.Ping‌ An. Low-bitrate Light Field Video Compression Through‌ Key Sequences Encoding and Joint Reconstruction Network.‌IEEE Transactions on Circuits and Systems for Video Technology2025, 1-1‌In press. HAL DOI‌
19 articleJ.Jingtao‌‌ Li, Z.Zhixuan Zhou, X.Xintong‌ Zhao, M.Mounir‌ Kaaniche, C.Congcong‌‌ Wang and D.Ding Liu. SU3plus: An‌ Enhanced Swin-UNet for Synthesizing‌ Vertically Integrated Liquid from‌‌ Multiple Meteorological Satellite Data.IEEE Transactions on‌ Geoscience and Remote Sensing‌2025, 1-1HAL‌‌DOI
20 articleS.Ségolène Martin, J.-C.‌Jean-Christophe Pesquet, G.‌Gabriele Steidl and I.‌‌ B.Ismail Ben Ayed. Variable Bregman majorization-minimization‌ algorithm and its application‌ to Dirichlet maximum likelihood‌‌ estimation.Journal of Applied and Numerical Optimization‌732025HAL‌DOI back to text‌‌
21 articleV.Vahan Martirosyan, J. H.‌Jhony H Giraldo and‌ F. D.Fragkiskos D‌‌ Malliaros. Piecewise Constant Spectral Graph Neural Network‌.Transactions on Machine‌ Learning Research JournalApril‌‌ 2025HAL back to text back to text‌
22 articleM.Malik‌ Nafa, L.Ludovic‌‌ Trautmann, U.Ugo Arles Bottega, T.‌Tigran Mansuryan, A.‌Alessandro Tonello, L.‌‌Lionel Rechignat, M.Marc Fabert, L.‌Laetitia Magnol, V.‌Véronique Blanquet, F.‌‌Fabienne Baraige, V.Vincent Couderc, C.‌Claire Carrion, J.-R.‌Jean-René Duclère and C.‌‌Claire Lefort. Label-free discrimination of muscle proteins‌ based on nonresonant background‌ signature of Multiplex-Coherent Anti-Stokes‌‌ Raman Scattering imaging.Journal of Physics D:‌ Applied Physics2025.‌ In press. HAL DOI‌‌back to text
23 articleJ.Joeri Nicolaes‌, E.Evi Tselenti‌, T.Theodore Aouad‌‌, C.Clementina López-Medina, A.Antoine Feydy‌, H.Hugues Talbot‌, B.Bengt Hoepken‌‌, N.Natasha de Peyrecave and M.Maxime‌ Dougados. Performance analysis‌ of a deep-learning algorithm‌‌ to detect the presence of inflammation in MRI‌ of sacroiliac joints in‌ patients with axial spondyloarthritis‌‌.Annals of the Rheumatic DiseasesJanuary 2025‌HAL DOI back to‌ text
24 articleN.‌‌Nora Ouzir, V.Vassili Pustovalov, D.-H.‌Duong-Hung Pham, D.‌Denis Kouamé and J.-C.‌‌Jean-Christophe Pesquet. A Proximal Algorithm for Joint‌ Blood Flow Computation and‌ Tissue Motion Compensation in‌‌ Doppler Ultrafast Ultrasound Imaging.Ultrasonics2025.‌ In press. HAL back‌ to text back to‌‌ text back to text
25 articleB.Barbara‌ Pistilli, F.Fernanda‌ Mosele, N.Noemie‌‌ Corcos, L.Livia Pierotti, Y.Yoann‌ Pradat, L.Loïc‌ Le Bescond, M.‌‌Magali Lacroix-Triki, G.Ghada Nachabeh, A.‌Alexia Alfaro, C.‌Cyril Catelain, B.‌‌Bastien Job, F.Fathia Mami-Chouaib, S.‌Severine Badel, F.‌Françoise Farace, M.‌‌Marianne Oulhen, P.Patricia Kannouche, D.‌ T.Diep T N‌ Tran, N.Nathalie‌‌ Droin, C.Cecile Vicier, J. S.‌Jean Sebastien Frenel,‌ V.Veronique D’hondt,‌‌ F.Florence Dalenc, T.Thomas Bachelot,‌ A.Agnes Ducoulombier,‌ M. A.Marc Antoine‌‌ Benderra, D.Delphine‌ Loirat, D.Didier Mayeur, E.Elise‌ Deluche, J.Jacqueline Deneuve, R.Rasha‌ Cheikh-Hussin, P.Pierre Guyader, N.Nicolas‌ Signolle, K.Karine Godefroy, H.Hugues‌ Talbot, M.Maria Vakalopoulou, S.Stergios‌ Christodoulidis, E.Elsa Bernard, Y.Yves‌ Koudou, A.Andrea Sporchia, F.Fumitaka‌ Suto, L.Lie Li, D. W.‌David W Sternberg, S.Stefan Michiels,‌ F.Fabrice André, D.Dalila Sellami and‌ G.Guillaume Montagnac. Patritumab deruxtecan in HR+HER2−‌ advanced breast cancer: a phase 2 trial.‌Nature MedicineSeptember 2025, Online ahead of‌ printHAL DOI back to text
26 article‌A. S.Amadou S. Sangare, N.Nicolas‌ Dunou, J. H.Jhony H. Giraldo and‌ F. D.Fragkiskos D. Malliaros. A Fused‌ Gromov-Wasserstein Approach to Subgraph Contrastive Learning.Transactions‌ on Machine Learning Research JournalFebruary 2025HAL‌back to text back to text
27 article‌A.Antoine Siraudin, F.Fragkiskos Malliaros and‌ C.Christopher Morris. Cometh: A continuous-time discrete-state‌ graph diffusion model.Transactions of Machine Learning‌ ResearchApril 2025HALback to text back‌ to text
28 articleM.Mathieu Vu,‌ E.Emilie Chouzenoux, I. B.Ismail Ben‌ Ayed and J.-C.Jean-Christophe Pesquet. Aggregated f‌ -average Neural Network applied to Few-Shot Class Incremental‌ Learning.Signal ProcessingJanuary 2025HAL back‌ to text

International peer-reviewed conferences

29 inproceedingsY.‌Yassine Abbahaddou, F. D.Fragkiskos D. Malliaros‌, J. F.Johannes F. Lutzeyer, A.‌ M.Amine Mohamed Aboussalah and M.Michalis Vazirgiannis‌. Graph Neural Network Generalization with Gaussian Mixture‌ Model Based Augmentation.Proceedings of the 42‌ nd International Conference on Machine Learning, Vancouver, Canada.‌ PMLR 267, 2025.ICML 2025 - 42nd International‌ Conference on Machine LearningVancoucer, Canada2025HAL‌back to text back to text
30 inproceedings‌Y.Yassine Abbahaddou, F. D.Fragkiskos D.‌ Malliaros, J. F.Johannes F. Lutzeyer and‌ M.Michalis Vazirgiannis. ADMP-GNN: Adaptive Depth Message‌ Passing GNN.CIKM 2025 - ACM International‌ Conference on Information and Knowledge ManagementSeoul, South‌ Korea2025, 4 - 13HAL DOI‌back to text back to text
31 inproceedings‌M.Miguel Amorim, J.-C.Jean-Christophe Pesquet,‌ T.Tristan Portugues, A.Antonio Silveti-Falls and‌ H.Hugues Talbot. Numerical Implementation of Variational‌ Morphological Operators on Graphs.DGMM 2025 -‌ 4th International Conference on Discrete Geometry and Mathematical‌ MorphologyGroningen, Netherlands, NetherlandsNovember 2025HAL back‌ to text
32 inproceedingsH.Hugo Attali,‌ T.Thomas Papastergiou, N.Nathalie Pernelle and‌ F. D.Fragkiskos D Malliaros. Dynamic Triangulation-Based‌ Graph Rewiring for Graph Neural Networks.CIKM‌ 2025 - ACM International Conference on Information and‌ Knowledge ManagementSeoul, South KoreaAugust 2025,‌ 87-97HAL DOI back to text back to‌ text
33 inproceedingsG.Gaspard Blaise, C.Clément Cosserat, E.‌Emilie Chouzenoux, J.-C.‌Jean-Christophe Pesquet and T.‌‌Tülay Adali. Deep Unrolled Architecture for Fast‌ and Accurate Gaussian Independent‌ Vector Analysis.SSP‌‌ 2025 - IEEE Statistical Signal Processing WorkshopEdimbourg‌ (Ecosse), United KingdomJune‌ 2025HAL back to‌‌ text
34 inproceedingsC.Clément Cosserat, L.‌Lucas Gois, E.‌Emilie Chouzenoux, V.‌‌Vince Calhoun and T.Tülay Adali. How‌ important are inter-dataset interactions‌ for large scale analysis‌‌ of fMRI data: A multi-dimensional comparison.Proceedings‌ of IEEE International Symposium‌ on Biomedical ImagingISBI‌‌ 2026 - IEEE International Symposium on Biomedical Imaging‌London, United KingdomApril‌ 2026HAL back to‌‌ text back to text
35 inproceedingsB.Benjamin‌ Cox, E.Emilie‌ Chouzenoux and V.Víctor‌‌ Elvira. Learning a sparse polynomial approximation to‌ the transition function of‌ general state-space models.‌‌Proceedings of the IEEE International Conference on Acoustics,‌ Speech and Signal Processing‌ (ICASSP 2025)ICASSP 2025‌‌ - IEEE International Conference on Acoustics, Speech and‌ Signal ProcessingHyderabad, India‌April 2025HAL back‌‌ to text back to text
36 inproceedingsA.‌Aref Einizade, D.‌Dorina Thanou, F.‌‌ D.Fragkiskos D Malliaros and J. H.Jhony‌ Heriberto Giraldo Zuluaga.‌ Continuous Simplicial Neural Networks‌‌.NeurIPS 2025 - 39th Annual Conference on‌ Neural Information Processing Systems‌San Diego, United States‌‌December 2025HAL back to text back to‌ text
37 inproceedingsH.‌Hafsa El Herichi,‌‌ A.Arturo Mendoza, Y.Yanneck Wielhorski,‌ H.Hugues Talbot and‌ S.Stéphane Roux.‌‌ Analyse de forme a priori pour la segmentation‌ des torons dans les‌ images tomographiques.Journées‌‌ Nationales sur les Composites 2025Gif-sur-Yvette, FranceJune‌ 2025HAL back to‌ text
38 inproceedings D.‌‌Daniele Malitesta, G.Giacomo Medda, E.‌Erasmo Purificato, M.‌Mirko Marras, F.‌‌ D.Fragkiskos D. Malliaros and L.Ludovico Boratto‌. How Fair is‌ Your Diffusion Recommender Model?‌‌ RecSys '25: Proceedings of the Nineteenth ACM Conference‌ on Recommender Systems RecSys‌ 2025 - 19th ACM‌‌ Conference on Recommender Systems Prague, Czech Republic September‌ 2025 HAL back to‌ text back to text‌‌
39 inproceedingsN.Nabil Mouadden, O.Othmane‌ Laousy, R.Rafael‌ Marini, V.Valentin‌‌ Ong, M.-P.Marie-Pierre Revel, G.Guillaume‌ Chassagnon, S.Stergios‌ Christodoulidis and M.Maria‌‌ Vakalopoulou. Conditional Latent Diffusion Models for Irregularly‌ Spaced Longitudinal Radiological Data‌.Medical Image Computing‌‌ and Computer Assisted Intervention – MICCAI 202515974‌Lecture Notes in Computer‌ ScienceLNCS-15974Daejeon, South‌‌ KoreaSpringer Nature SwitzerlandSeptember 2026, 106-115‌HAL DOI back to‌ text
40 inproceedingsT.‌‌Thomas Pethick, W.Wanyun Xie, K.‌Kimon Antonakopoulos, Z.‌Zhenyu Zhu, A.‌‌Antonio Silveti-Falls and V.Volkan Cevher. Training‌ Deep Learning Models with‌ Norm-Constrained LMOs.42nd‌‌ International Conference on Machine LearningVancouver, CanadaJuly‌ 2025HAL back to‌ text back to text‌‌back to text
41‌ inproceedingsT.Thomas Pethick, W.Wanyun Xie‌, M.Mete Erdogan, K.Kimon Antonakopoulos‌, T.Tony Silveti-Falls and V.Volkan Cevher‌. Generalized Gradient Norm Clipping & Non-Euclidean (L‌ 0 , L 1 )-Smoothness.NeurIPS 2025‌ - 39th Conference on Neural Information Processing Systems‌San Diego (CA), United StatesDecember 2025HAL‌back to text back to text back to‌ text
42 inproceedingsD.David Restrepo, I.‌Ira Ktena, M.Maria Vakalopoulou, S.‌Stergios Christodoulidis and E.Enzo Ferrante. On‌ the Risk of Misleading Reports: Diagnosing Textual Biases‌ in Multimodal Clinical AI.AI for Clinical‌ Applications. Agentic AI16147Lecture Notes in Computer‌ ScienceDaejean, South KoreaSpringer Nature SwitzerlandSeptember‌ 2026, 320-330HALDOI back to text‌
43 inproceedingsT.Théo Sourget, D.David‌ Restrepo, C.Céline Hudelot, E.Enzo‌ Ferrante, S.Stergios Christodoulidis and M.Maria‌ Vakalopoulou. Fairness and Robustness of CLIP-Based Models‌ for Chest X-Rays.Fairness of AI in‌ Medical Imaging. FAIMI 202515976Lecture Notes in‌ Computer ScienceDaejean, South KoreaSpringer Nature Switzerland‌September 2026, 11-21HAL DOI back to‌ text back to text
44 inproceedingsM.Mingbo‌ Su, Y.Yi Zhang, M.Mounir‌ Kaaniche, C.Congcong Wang and M.Meng‌ Zhao. SR-MedTS: A Semantically-Robust Synergistic Framework for‌ multi-modal Medical Image Translation and Segmentation.BIBM‌ 2025 - International Conference on Bioinformatics and Biomedicine‌Wuhan, ChinaDecember 2025HAL
45 inproceedingsK.‌Kazuki Suehara, Y.Yi Gu, Y.‌Yoshito Otake, K.Keisuke Uemura, M.‌Masashi Okamoto, K.Kunihiko Tokunaga, H.‌Hugues Talbot and Y.Yoshinobu Sato. Estimating‌ Bone Mineral Density and Muscle Mass from EOS‌ Low Dose X-Ray Imaging System.Lecture Notes‌ in Computer Science (LNCS, volume 15970)Medical Image‌ Computing and Computer Assisted Intervention (MICCAI)15970Lecture‌ Notes in Computer ScienceDaejong, South KoreaSpringer‌ Nature SwitzerlandSeptember 2026, 168-177HAL DOI‌back to text
46 inproceedingsL.Long Zhou‌, F.Fereshteh Shakeri, A.Aymen Sadraoui‌, M.Mounir Kaaniche, J.-C.Jean-Christophe Pesquet‌ and I. B.Ismail Ben Ayed. UNEM:‌ UNrolled Generalized EM for Transductive Few-Shot Learning.‌CVPR 2025 - Computer Vision and Pattern Recognition‌ ConferenceNashville (Tenessee), United StatesJune 2025HAL‌back to text

Conferences without proceedings

47 inproceedings‌M.Malik Nafa, L.Ludovic Trautmann,‌ U.Ugo Arles Bottega, T.Tigran Mansuryan‌, A.Alessandro Tonello, L.Lionel Rechignat‌, M.Marc Fabert, L.Laetitia Magnol‌, V.Véronique Blanquet, F.Fabienne Baraige‌, V.Vincent Couderc, C.Claire Carrion‌, J.-R.Jean-René Duclère and C.Claire Lefort‌. Imagerie CARS multiplexée pour la discrimination sans‌ marquage des protéines musculaires.IMOTOP 2025 -‌ Imagerie et résolution ultimeVillneuve d'Ascq, FranceOctober‌ 2025HAL

Reports & preprints

48 miscA.Abbas Khademi and A.‌Antonio Silveti-Falls. Adaptive‌ Conditional Gradient Descent.‌‌October 2025HAL
49 miscL.Loïc Le‌ Bescond, M.Marvin‌ Lerousseau, F.Fabrice‌‌ Andre and H.Hugues Talbot. SparseXMIL: Leveraging‌ sparse convolutions for context-aware‌ and memory-efficient classification of‌‌ whole slide images in digital pathology.January‌ 2025HAL back to‌ text
50 miscS.‌‌Solal Martin, E.Emilie Chouzenoux and V.‌Víctor Elvira. A‌ novel sequential method for‌‌ building upper and lower bounds of moments of‌ distributions.December 2025‌, 41 p.HAL‌‌
51 miscT.Thomas Pethick, K.Kimon‌ Antonakopoulos, A.Antonio‌ Silveti-Falls, L. C.‌‌Leena Chennuru Vankadara and V.Volkan Cevher.‌ Training Neural Networks at‌ Any Scale.December‌‌ 2025HAL back to text
52 miscN.‌Nicolas Salvy, H.‌Hugues Talbot and B.‌‌Bertrand Thirion. Enhanced generative model evaluation with‌ Clipped Density and Coverage‌.July 2025HAL‌‌back to text
53 miscF.Francesco Songia‌, R.Raoul Sallé‌ de Chou, H.‌‌Hugues Talbot and I.Irene Vignon-Clementel. Multi-fidelity‌ graph-based neural networks architectures‌ to learn Navier-Stokes solutions‌‌ on non-parametrized 2D domains.December 2025HAL‌

OPIS - 2025

OPIS - 2025

2025Activity​​​‌ reportProject-TeamOPIS

Keywords

Computer﻿​​﻿ Science and Digital Science​​​‌

Other Research Topics﻿‌​‌ and Application Domains

1 Team﻿​​﻿ members, visitors, external collaborators​​​‌

Research Scientist

Faculty​​​‌ Members

Post-Doctoral Fellows﻿‌​‌

PhD Students

Technical Staff

Interns​​​‌ and Apprentices

Administrative Assistants​​﻿﻿

Visiting Scientists

2 Overall objectives

3 Research program﻿﻿﻿‌

3.1 Accelerated algorithms for﻿‌​‌ solving high-dimensional continuous optimization﻿​​﻿ problems

3.2 Optimization over graphs​‌﻿﻿

3.3 Toward​​﻿﻿ more understandable deep learning​​​‌

4 Application domains﻿﻿﻿‌

4.1 Robustness of AI﻿‌​‌

4.1.1 Robust few-shot learning﻿​​﻿

4.1.2﻿‌​‌ Design of robust neural﻿​​﻿ networks in safety critical​​​‌ industrial domains

4.1.3​​﻿﻿ Certification of segmentation networks​​​‌

4.2 Radiology,​‌﻿﻿ hyper-progressive disease and immunotherapy​​﻿﻿

4.2.1 Realtime flow estimation​​​‌ for interventional doppler ultrafast﻿​﻿﻿ ultrasound

4.2.2 Imaging radiomics​‌﻿﻿ and pathomics to assess​​﻿﻿ response to treatment

Participants:​​​‌

4.2.3 Analysis of﻿​​﻿ histopathology images for cancer​​​‌ treatment

4.2.4 Vision,​​​‌ machine learning and precision﻿﻿﻿‌ medicine

4.2.5﻿﻿﻿‌ Physics-informed, generative models for﻿‌​‌ heart left ventricle perfusion﻿​​﻿ analysis

4.3﻿‌​‌ Sparse inverse problems

4.3.1﻿​​﻿ Sparse signal processing in​​​‌ chemistry

4.3.2﻿﻿﻿‌ Image restoration for Multiplex-Coherent﻿‌​‌ Anti-Stokes Raman Scattering (M-CARS)﻿​​﻿ spectroscopy

4.3.3 Reconstruction approaches in﻿​​﻿ PET imaging

4.4﻿​​﻿ Graph mining applications

4.4.1​​​‌ Geometric Graph Neural Networks﻿﻿﻿‌ for molecular and chemical﻿‌​‌ systems

4.4.2 Graph machine​‌﻿﻿ learning for spatiotemporal data​​﻿﻿

4.4.3 Graph​‌﻿﻿ representation learning for computational​​﻿﻿ medicine

4.4.4 Graph-based models for​‌﻿﻿ multimodal patient representation learning​​﻿﻿

4.4.5 Graph​​​‌ inference for time series﻿﻿﻿‌ analysis in complex dynamical﻿‌​‌ systems

4.5 Other﻿​​﻿ biomedical applications

4.5.1 Visual​​​‌ Language Foundation Models for﻿﻿﻿‌ medical image diagnosis

4.5.2​​​‌ Independant vector analysis for﻿﻿﻿‌ fMRI data processing

4.5.3﻿﻿﻿‌ Imaging biomarkers and characterization﻿‌​‌ for chronic lung diseases﻿​​﻿

4.5.4 AI​​​‌ for small bowel obstruction﻿​﻿﻿ diagnosis

4.5.5 A generative model​​﻿﻿ for heart left ventricle​​​‌ perfusion analysis

5​​​‌ Social and environmental responsibility﻿﻿﻿‌

5.1 Footprint of research﻿‌​‌ activities

5.2 Impact of research﻿﻿﻿‌ results

6 Highlights﻿‌​‌ of the year

6.1﻿​​﻿ Awards

7 Latest software​​​‌ developments, platforms, open data﻿​﻿﻿

7.1 New platforms

7.1.1​‌﻿﻿ Prox Repository

7.1.2 The​​﻿﻿ PINK image library

7.1.3 The﻿​﻿﻿ Vivabrain AngioTK toolkit

7.1.4﻿​﻿﻿ A scientific image viewer​‌﻿﻿

7.1.5​​​‌ Scion

7.1.6 ScanCovIA

7.1.7 FOSSIL

7.1.8﻿﻿﻿‌ PieCoN

7.1.9 Cometh﻿‌​‌

7.1.10​‌﻿﻿ Gratin

7.1.11 FairDiffRec﻿​﻿﻿

7.1.12 ADMP-GNN

7.1.13 TRIGON

7.1.14 COSIMO

8 New results​​﻿﻿

8.1 Deep unrolled algorithms​​​‌ for inverse problems in﻿​﻿﻿ image and signal processing​‌﻿﻿

8.2 Unrolled﻿﻿﻿‌ Generalized EM for Transductive﻿‌​‌ Few-Shot Learning

8.3 Learning﻿﻿﻿‌ truly monotone operators

8.4 Majorization-Minimization algorithm and​​​‌ application to Dirichlet maximum﻿​﻿﻿ likelihood estimation

8.5 Multi-task​‌﻿﻿ neural networks for lifting-based​​﻿﻿ image coders

8.6﻿‌​‌ Training Neural Networks with﻿​​﻿ non-Euclidean geometry and spectral​​​‌ constraints

8.7 Variational morphological operators​​​‌ on graphs

8.8 Segmentation of​​​‌ fiber bundles in tomography﻿﻿﻿‌

8.9 Convergence analysis in​​​‌ stochastic optimization

8.10 Divergence minimization​​​‌ in statistical inference

2025Activity‌ reportProject-TeamOPIS

Computer Science and Digital Science‌

Other Research Topics‌‌ and Application Domains

1 Team members, visitors, external collaborators‌

Faculty‌ Members

Post-Doctoral Fellows‌‌

Interns‌ and Apprentices

Administrative Assistants

3 Research program‌

3.1 Accelerated algorithms for‌‌ solving high-dimensional continuous optimization problems

3.2 Optimization over graphs‌

3.3 Toward more understandable deep learning‌

4 Application domains‌

4.1 Robustness of AI‌‌

4.1.1 Robust few-shot learning

4.1.2‌‌ Design of robust neural networks in safety critical‌ industrial domains

4.1.3 Certification of segmentation networks‌

4.2 Radiology,‌ hyper-progressive disease and immunotherapy

4.2.1 Realtime flow estimation‌ for interventional doppler ultrafast ultrasound

4.2.2 Imaging radiomics‌ and pathomics to assess response to treatment

Participants:‌

4.2.3 Analysis of histopathology images for cancer‌ treatment

4.2.4 Vision,‌ machine learning and precision‌ medicine

4.2.5‌ Physics-informed, generative models for‌‌ heart left ventricle perfusion analysis

4.3‌‌ Sparse inverse problems

4.3.1 Sparse signal processing in‌ chemistry

4.3.2‌ Image restoration for Multiplex-Coherent‌‌ Anti-Stokes Raman Scattering (M-CARS) spectroscopy

4.3.3 Reconstruction approaches in PET imaging

4.4 Graph mining applications

4.4.1‌ Geometric Graph Neural Networks‌ for molecular and chemical‌‌ systems

4.4.2 Graph machine‌ learning for spatiotemporal data

4.4.3 Graph‌ representation learning for computational medicine

4.4.4 Graph-based models for‌ multimodal patient representation learning

4.4.5 Graph‌ inference for time series‌ analysis in complex dynamical‌‌ systems

4.5 Other biomedical applications

4.5.1 Visual‌ Language Foundation Models for‌ medical image diagnosis

4.5.2‌ Independant vector analysis for‌ fMRI data processing

4.5.3‌ Imaging biomarkers and characterization‌‌ for chronic lung diseases

4.5.4 AI‌ for small bowel obstruction diagnosis

4.5.5 A generative model for heart left ventricle‌ perfusion analysis

5‌ Social and environmental responsibility‌

5.1 Footprint of research‌‌ activities

5.2 Impact of research‌ results

6 Highlights‌‌ of the year

6.1 Awards

7 Latest software‌ developments, platforms, open data

7.1.1‌ Prox Repository

7.1.2 The PINK image library

7.1.3 The Vivabrain AngioTK toolkit

7.1.4 A scientific image viewer‌

7.1.5‌ Scion

7.1.8‌ PieCoN

7.1.9 Cometh‌‌

7.1.10‌ Gratin

7.1.11 FairDiffRec

8 New results

8.1 Deep unrolled algorithms‌ for inverse problems in image and signal processing‌

8.2 Unrolled‌ Generalized EM for Transductive‌‌ Few-Shot Learning

8.3 Learning‌ truly monotone operators

8.4 Majorization-Minimization algorithm and‌ application to Dirichlet maximum likelihood estimation

8.5 Multi-task‌ neural networks for lifting-based image coders

8.6‌‌ Training Neural Networks with non-Euclidean geometry and spectral‌ constraints

8.7 Variational morphological operators‌ on graphs

8.8 Segmentation of‌ fiber bundles in tomography‌

8.9 Convergence analysis in‌ stochastic optimization

8.10 Divergence minimization‌ in statistical inference

8.11 Fast schemes for adaptive importance‌ sampling

8.12 Computational approaches for drug discovery

8.13‌ Analysis of pathology whole slide images with spatial‌ context

8.14 Robust‌ automatic crater detection at‌‌ all latitudes on Mars with Deep-learning

8.15 Expectation-Maximization for time series modeling and‌ inference

8.16 Interpretable ensembling rules‌

8.17 Boosting radiologist performance‌ using machine learning, the usecase of parotid tumours‌ diagnosis

8.18 Iterative optimization for independent vector analysis

8.19 Proximal algorithm‌ for joint blood flow‌‌ computation and tissue motion compensation in Doppler ultrafast‌ ultrasound imaging

8.20 A Fused‌ Gromov-Wasserstein Approach to Subgraph‌ Contrastive Learning

8.21 Piecewise‌ Constant Spectral Graph Neural‌‌ Network

8.22 Cometh: A‌ Continuous-time Discrete-state Graph Diffusion Model

8.23 Graph Neural Network‌ Generalization With Gaussian Mixture Model Based Augmentation

8.24 How‌ Fair is Your Diffusion Recommender Model?