2025Activity​ reportProject-TeamPOLARIS

Project-team positioning

Evaluating​​​‌ the scalability, robustness, energy‌ consumption and performance of‌​‌ large infrastructures such as​​ exascale platforms and clouds​​​‌ raises severe methodological challenges.‌ The complexity of such‌​‌ platforms mandates empirical evaluation​​ but direct experimentation via​​​‌ an application deployment on‌ a real-world testbed is‌​‌ often limited by the​​ few platforms available at​​​‌ hand and is even‌ sometimes impossible (cost, access,‌​‌ early stages of the​​ infrastructure design, etc.). Furthermore,​​​‌ such experiments are costly,‌ difficult to control and‌​‌ therefore difficult to reproduce.​​ Although many of these​​​‌ digital systems have been‌ built by human, they‌​‌ have reached such a​​ complexity level that we​​​‌ are no longer able‌ to study them like‌​‌ artificial systems and have​​ to deal with the​​​‌ same kind of experimental‌ issues as natural sciences.‌​‌ The development of a​​ sound experimental methodology for​​​‌ the evaluation of resource‌ management solutions is among‌​‌ the most important ways​​ to cope with the​​​‌ growing complexity of computing‌ environments. Although computing environments‌​‌ come with their own​​ specific challenges, we believe​​​‌ such general observation problems‌ should be addressed by‌​‌ borrowing good practices and​​ techniques developed in many​​​‌ other domains of science,‌ in particular (1) Predictive‌​‌ Simulation, (2) Trace Analysis​​ and Visualization, and (3)​​​‌ the Design of Experiments.‌

Scientific achievements

Large computing‌​‌ systems are particularly complex​​ to understand because of​​​‌ the interplay between their‌ discrete nature (originating from‌​‌ deterministic computer programs) and​​ their stochastic nature (emerging​​​‌ from the physical world,‌ long distance interactions, and‌​‌ complex hardware and software​​ stacks). A first line​​​‌ of research in POLARIS‌ is devoted to the‌​‌ design of relatively simple​​ statistical models of key​​​‌ components of distributed systems‌ and their exploitation to‌​‌ feed simulators of such​​ systems, to build aggregated​​​‌ performance views, and to‌ detect anomalous behaviors.

Project-team​​ positioning

Stochastic models often​​​‌ suffer from the curse‌ of dimensionality: their complexity‌​‌ grows exponentially with the​​ number of dimensions of​​​‌ the system. At the‌ same time, very large‌​‌ stochastic systems are sometimes​​ easier to analyze: it​​​‌ can be shown that‌ some classes of stochastic‌​‌ systems simplify as their​​ dimension goes to infinity​​​‌ because of averaging effects‌ such as the law‌​‌ of large numbers, or​​ the central limit theorem.​​​‌ This forms the basis‌ of what is called‌​‌ an asymptotic method,​​​‌ which consists in studying​ what happens when a​‌ system gets large in​​ order to build an​​​‌ approximation that is easier​ to study or to​‌ simulate.

Within the team,​​ the research that we​​​‌ conduct in this axis​ is to foster the​‌ applicability of these asymptotic​​ methods to new application​​​‌ areas. This leads us​ to work on the​‌ application of classical methods​​ to new problems, but​​​‌ also to develop new​ approximation methods that take​‌ into account special features​​ of the systems we​​​‌ study (i.e., moderate number​ of dimensions, transient behavior,​‌ random matrices). Typical applications​​ are mean field method​​​‌ for performance evaluation, application​ to distributed optimization, and​‌ more recently statistical learning.​​ One originality of our​​​‌ work is to quantify​ precisely what is the​‌ error made by such​​ approximations. This allows us​​​‌ to define refinement terms​ that lead to more​‌ accurate approximations.

Scientific achievements​​

Project-team positioning​​​‌

Online learning concerns the​ study of repeated decision-making​‌ in changing environments. Of​​ course, depending on the​​​‌ context, the words “learning”​ and “decision-making” may refer​‌ to very different things:​​ in economics, this could​​​‌ mean predicting how rational​ agents react to market​‌ drifts; in data networks,​​ it could mean adapting​​​‌ the way packets are​ routed based on changing​‌ traffic conditions; in machine​​ learning and AI applications,​​​‌ it could mean training​ a neural network or​‌ the guidance system of​​ a self-driving car; etc.​​​‌ In particular, the changes​ in the learner's environment​‌ could be either exogenous​​ (that is, independent of​​​‌ the learner's decisions, such​ as the weather affecting​‌ the time of travel),​​ or endogenous (i.e., they​​​‌ could depend on the​ learner's decisions, as in​‌ a game of poker),​​ or any combination thereof.​​​‌ However, the goal for​ the learner(s) is always​‌ the same: to make​​ more informed decisions that​​​‌ lead to better rewards​ over time.

The study​‌ of online learning models​​ and algorithms dates back​​​‌ to the seminal work​ of Robbins, Nash and​‌ Bellman in the 50's,​​ and it has since​​​‌ given rise to a​ vigorous research field at​‌ the interface of game​​ theory, control and optimization,​​​‌ with numerous applications in​ operations research, machine learning,​‌ and data science. In​​ this general context, our​​​‌ team focuses on the​ asymptotic behavior of online​‌ learning and optimization algorithms,​​ both single- and multi-agent:​​​‌ whether they converge, at​ what speed, and/or what​‌ type of non-stationary, off-equilibrium​​ behaviors may arise when​​​‌ they do not.

The​ focus of POLARIS on​‌ game-theoretic and Markovian models​​ of learning covers a​​​‌ set of specific challenges​ that dovetail in a​‌ highly synergistic manner with​​ the work of other​​​‌ learning-oriented teams within Inria​ (like SCOOL in Lille,​‌ SIERRA in Paris, and​​ THOTH in Grenoble), and​​​‌ it is an important​ component of Inria's activities​‌ and contributions in the​​ field (which includes major​​​‌ industrial stakeholders like Google​ / DeepMind, Facebook, Microsoft,​‌ Amazon, and many others).​​

Scientific achievements

Our team's​​​‌ work on online learning​ covers both single- and​‌ multi-agent models; in the​​ sequel, we present some​​​‌ highlights of our work​ structured along these basic​‌ axes.

In the single-agent​​ setting, an important problem​​​‌ in the theory of​ Markov decision processes –​‌ i.e., discrete-time control processes​​ with decision-dependent randomness –​​​‌ is the so-called “restless​ bandit” problem. Here, the​‌ learner chooses an action​​ – or “arm” –​​​‌ from a finite set,​ and the mechanism determining​‌ the action's reward changes​​ depending on whether the​​​‌ action was chosen or​ not (in contrast to​‌ standard Markov problems where​​ the activation of an​​​‌ arm does not have​ this effect). In this​‌ general setting, Whittle conjectured​​ – and Weber and​​​‌ Weiss proved – that​ Whittle's eponymous index policy​‌ is asymptotically optimal. However,​​ the result of Weber​​​‌ and Weiss is purely​ asymptotic, and the rate​‌ of this convergence remained​​ elusive for several decades.​​​‌ This gap was finally​ settled in a series​‌ of POLARIS papers 66​​67, where we​​ showed that Whittle indices​​​‌ (as well as other‌ index policies) become optimal‌​‌ at a geometric rate​​ under the same technical​​​‌ conditions used by Weber‌ and Weiss to prove‌​‌ Whittle's conjecture, plus a​​ technical requirement on the​​​‌ non-singularity of the fixed‌ point of the mean-field‌​‌ dynamics. We also propose​​ the first sub-cubic algorithm​​​‌ to compute Whittle and‌ Gittins indexes. As for‌​‌ reinforcement learning in Markovian​​ bandits, we have shown​​​‌ that Bayesian and optimistic‌ approaches do not use‌​‌ the structure of Markovian​​ bandits similarly: While Bayesian​​​‌ learning has both a‌ regret and a computational‌​‌ complexity that scales linearly​​ with the number of​​​‌ arms, optimistic approaches all‌ incur an exponential computation‌​‌ time, at least in​​ their current versions 65​​​‌.

In the multi-agent‌ setting, our work has‌​‌ focused on the following​​ fundamental question:

Does the​​​‌ concurrent use of (‌possibly optimal) single-agent‌​‌ learning algorithms

ensure convergence​​ to Nash equilibrium in​​​‌ multi-agent, game-theoretic environments?

Conventional‌ wisdom might suggest a‌​‌ positive answer to this​​ question because of the​​​‌ following “folk theorem”: under‌ no-regret learning, the agents'‌​‌ empirical frequency of play​​ converges to the game's​​​‌ set of coarse correlated‌ equilibria. However, the actual‌​‌ implications of this result​​ are quite weak: First,​​​‌ it concerns the empirical‌ frequency of play and‌​‌ not the day-to-day sequence​​ of actions employed by​​​‌ the players. Second, it‌ concerns coarse correlated equilibria‌​‌ which may be supported​​ on strictly dominated strategies​​​‌ – and are thus‌ unacceptable in terms of‌​‌ rationalizability. These realizations prompted​​ us to make a​​​‌ clean break with conventional‌ wisdom on this topic,‌​‌ ultimately showing that the​​ answer to the above​​​‌ question is, in general,‌ “no”: specifically, 93,‌​‌ 91 showed that the​​ (optimal) class of “follow-the-regularized-leader”​​​‌ (FTRL) learning algorithms leads‌ to Poincaré recurrence even‌​‌ in simple, $2\times 2$ min-max games, thus​​​‌ precluding convergence to Nash‌ equilibrium in this context.‌​‌

This negative result generated​​ significant interest in the​​​‌ literature as it contributed‌ in shifting the focus‌​‌ towards identifying which Nash​​ equilibria may arise as​​​‌ stable limit points of‌ FTRL algorithms and dynamics.‌​‌ Earlier work by POLARIS​​ on the topic 49​​​‌, 94, 95‌ suggested that strict Nash‌​‌ equilibria play an important​​ role in this question.​​​‌ This suspicion was recently‌ confirmed in a series‌​‌ of papers 61,​​ 78 where we established​​​‌ a sweeping negative result‌ to the effect that‌​‌ mixed Nash equilibria are​​ incompatible with no-regret learning​​​‌. Specifically, we showed‌ that any Nash equilibrium‌​‌ which is not strict​​ cannot be stable and​​​‌ attracting under the dynamics‌ of FTRL, especially in‌​‌ the presence of randomness​​ and uncertainty. This result​​​‌ has significant implications for‌ predicting the outcome of‌​‌ a multi-agent learning process​​ because, combined with 94​​​‌, it establishes the‌ following far-reaching equivalence: a‌​‌ state is asymptotically stable​​ under no-regret learning if​​​‌ and only if it‌ is a strict Nash‌​‌ equilibrium.

Going beyond finite​​ games, this further raised​​​‌ the question of what‌ type of non-convergent behaviors‌​‌ can be observed in​​​‌ continuous games – such​ as the class of​‌ stochastic min-max problems that​​ are typically associated to​​​‌ generative adversarial networks (GANs)​ in machine learning. This​‌ question was one of​​ our primary collaboration axes​​​‌ with EPFL, and led​ to a joint research​‌ project focused on the​​ characterization of the convergence​​​‌ properties of zeroth-, first-,​ and (scalable) second-order methods​‌ in non-convex/non-concave problems. In​​ particular, we showed in​​​‌ 82 that these state-of-the-art​ min-max optimization algorithms may​‌ converge with arbitrarily high​​ probability to attractors that​​​‌ are in no way​ min-max optimal or even​‌ stationary – and, in​​ fact, may not even​​​‌ contain a single stationary​ point (let alone a​‌ Nash equilibrium). Spurious convergence​​ phenomena of this type​​​‌ can arise even in​ two-dimensional problems, a fact​‌ which corroborates the empirical​​ evidence surrounding the formidable​​​‌ difficulty of training GANs.​

Project-team positioning​​​‌

The topics in this​ axis emerge from current​‌ social and economic questions​​ rather than from a​​​‌ fixed set of mathematical​ methods. To this end​‌ we have identified large​​ trends such as energy​​​‌ efficiency, fairness, privacy, and​ the growing number of​‌ new market places. In​​ addition, COVID has posed​​​‌ new questions that opened​ new paths of research​‌ with strong links to​​ policy making.

Throughout these​​​‌ works, the focus of​ the team is on​‌ modeling aspects of the​​ aforementioned problems, and obtaining​​​‌ strong theoretical results that​ can give high-level guidelines​‌ on the design of​​ markets or of decision-making​​​‌ procedures. Where relevant, we​ complement those works by​‌ measurement studies and audits​​ of existing systems that​​​‌ allow identifying key issues.​ As this work is​‌ driven by topics, rather​​ than methods, it allows​​​‌ for a wide range​ of collaborations, including with​‌ enterprises (e.g., Naverlabs), policy​​ makers, and academics from​​​‌ various fields (economics, policy,​ epidemiology, etc.).

Other teams​‌ at Inria cover some​​ of the societal challenges​​​‌ listed here (e.g., PRIVATICS,​ COMETE) but rather in​‌ isolation. The specificity of​​ POLARIS resides in the​​​‌ breadth of societal topics​ covered and of the​‌ collaborations with non-CS researchers​​ and non-research bodies; as​​​‌ well as in the​ application of methods such​‌ as game theory to​​ those topics.

Scientific achievements​​​‌

7.1.1 SimGrid

• Keywords:
  Large-scale​​​‌ Emulators, Grid Computing, Distributed‌ Applications
• Scientific Description:

  SimGrid‌​‌ is a toolkit that​​ provides core functionalities for​​​‌ the simulation of distributed‌ applications in heterogeneous distributed‌​‌ environments. The simulation engine​​ uses algorithmic and implementation​​​‌ techniques toward the fast‌ simulation of large systems‌​‌ on a single machine.​​ The models are theoretically​​​‌ grounded and experimentally validated.‌ The results are reproducible,‌​‌ enabling better scientific practices.​​

  Its models of networks,​​​‌ cpus and disks are‌ adapted to (Data)Grids, P2P,‌​‌ Clouds, Fogs, Clusters and​​ HPC, allowing multi-domain studies.​​​‌ It can be used‌ either to simulate algorithms‌​‌ and prototypes of applications,​​ or to emulate real​​​‌ MPI applications through the‌ virtualization of their communication,‌​‌ or to formally assess​​ algorithms and applications that​​​‌ can run in the‌ framework.

  The formal verification‌​‌ module explores all possible​​ message interleavings in the​​​‌ application, searching for states‌ violating the provided properties.‌​‌ This tool can be​​ used to assess safety​​​‌ properties over arbitrary and‌ legacy codes, thanks to‌​‌ a system-level introspection tool​​​‌ that provides a finely​ detailed view of the​‌ running application to the​​ model checker. This can​​​‌ for example be leveraged​ to verify arbitrary MPI​‌ code written in C/C++/Fortran.​​

• Functional Description:
  SimGrid is​​​‌ a simulation toolkit that​ provides core functionalities for​‌ the simulation of distributed​​ applications in large scale​​​‌ heterogeneous distributed environments.
• Release​ Contributions:

  Breaking the seal:​‌ v4.0 was not the​​ final release.

  * Allow​​​‌ one to unseal netzones​ to modify the platform​‌ even after the simulation​​ start. * The model-checker​​​‌ can now report memory​ race conditions (see tutorial).​‌ * Pip builds should​​ now work out of​​​‌ the box. * (+​ the usual bug fixes​‌ overall, and improvements to​​ the Java/Python bindings).

• News​​​‌ of the Year:

  There​ were 2 major releases​‌ in 2025. We released​​ v4.0 in March, embodying​​​‌ 10 years of development.​ This turns SimGrid into​‌ a mature and stable​​ research instrument. The users​​​‌ can easily extend this​ tool to adapt it​‌ to their specific research,​​ while trusting its software​​​‌ implementation. Release v4.1 was​ published in November, showing​‌ that the development did​​ not stall even if​​​‌ the framework is mostly​ in maintenance mode. The​‌ performance simulation mode was​​ extended to allow modifications​​​‌ of the platform topology​ during the simulation.

  Most​‌ of our work this​​ year is related to​​​‌ the use of SimGrid​ for performance prediction of​‌ HPC applications and capacity​​ planning of supercomputers.

• URL:​​​‌
  https://­simgrid.­org/
• Publication:
  hal-04909441
• Contact:​
  Martin Quinson
• Participants:
  Mathieu​‌ Laurent, Anne-Cécile Orgerie, Arnaud​​ Legrand, Augustin Degomme, Arnaud​​​‌ Giersch, Frédéric Suter, Martin​ Quinson, Samuel Thibault
• Partners:​‌
  CNRS, ENS Rennes

7.1.2​​ PSI

• Name:
  Perfect Simulator​​​‌
• Keywords:
  Markov model, Simulation​
• Functional Description:
  Perfect simulator​‌ is a simulation software​​ of markovian models. It​​​‌ is able to simulate​ discrete and continuous time​‌ models to provide a​​ perfect sampling of the​​​‌ stationary distribution or directly​ a sampling a functional​‌ of this distribution by​​ using coupling from the​​​‌ past. The simulation kernel​ is based on the​‌ CFTP algorithm, and the​​ internal simulation of transitions​​​‌ on the Aliasing method.​
• URL:
  https://­gitlab.­inria.­fr/­PSI/­psi3/
• Contact:
  Jean-Marc​‌ Vincent

7.1.3 marmoteCore

• Name:​​
  Markov Modeling Tools and​​​‌ Environments - the Core​
• Keywords:
  Modeling, Stochastic models,​‌ Markov model
• Functional Description:​​

  marmoteCore is a C++​​​‌ environment for modeling with​ Markov chains. It consists​‌ in a reduced set​​ of high-level abstractions for​​​‌ constructing state spaces, transition​ structures and Markov chains​‌ (discrete-time and continuous-time). It​​ provides the ability of​​​‌ constructing hierarchies of Markov​ models, from the most​‌ general to the particular,​​ and equip each level​​​‌ with specifically optimized solution​ methods.

  This software was​‌ started within the ANR​​ MARMOTE project: ANR-12-MONU-00019.

• URL:​​​‌
  https://­gitlab.­inria.­fr/­PSI/­marmotecore/
• Publications:
  hal-01651940,​ hal-01276456
• Contact:
  Alain Jean-Marie​‌
• Participants:
  Alain Jean-Marie, Hlib​​ Mykhailenko, Benjamin Briot, Franck​​​‌ Quessette, Issam Rabhi, Jean-Marc​ Vincent, Jean-Michel Fourneau
• Partners:​‌
  Université de Versailles St-Quentin-en-Yvelines,​​ Université Paris Nanterre

7.1.4​​​‌ MarTO

• Name:
  Markov Toolkit​ for Markov models simulation:​‌ perfect sampling and Monte​​ Carlo simulation
• Keywords:
  Perfect​​​‌ sampling, Markov model
• Functional​ Description:
  MarTO is a​‌ simulation software of markovian​​ models. It is able​​ to simulate discrete and​​​‌ continuous time models to‌ provide a perfect sampling‌​‌ of the stationary distribution​​ or directly a sampling​​​‌ of functional of this‌ distribution by using coupling‌​‌ from the past. The​​ simulation kernel is based​​​‌ on the CFTP algorithm,‌ and the internal simulation‌​‌ of transitions on the​​ Aliasing method. This software​​​‌ is a rewrite, more‌ efficient and flexible, of‌​‌ PSI
• URL:
  https://­gitlab.­inria.­fr/­MarTo/­marto
• Contact:​​
  Vincent Danjean

7.1.5 GameSeer​​​‌

• Keyword:
  Game theory
• Functional‌ Description:
  GameSeer is a‌​‌ tool for students and​​ researchers in game theory​​​‌ that uses Mathematica to‌ generate phase portraits for‌​‌ normal form games under​​ a variety of (user-customizable)​​​‌ evolutionary dynamics. The whole‌ point behind GameSeer is‌​‌ to provide a dynamic​​ graphical interface that allows​​​‌ the user to employ‌ Mathematica's vast numerical capabilities‌​‌ from a simple and​​ intuitive front-end. So, even​​​‌ if you've never used‌ Mathematica before, you should‌​‌ be able to generate​​ fully editable and customizable​​​‌ portraits quickly and painlessly.‌
• URL:
  http://­polaris.­imag.­fr/­panayotis.­mertikopoulos/­publications/­s01-gameseer/
• Contact:
  Panayotis‌​‌ Mertikopoulos

7.1.6 rmf_tool

• Name:​​
  A library to Compute​​​‌ (Refined) Mean Field Approximations‌
• Keyword:
  Mean Field
• Functional‌​‌ Description:

  The tool accepts​​ three model types:

  -​​​‌ homogeneous population processes (HomPP)‌

  - density dependent population‌​‌ processes (DDPPs)

  - heterogeneous​​ population models (HetPP)

  In​​​‌ particular, it provides a‌ numerical algorithm to compute‌​‌ the constant of the​​ refined mean field approximation​​​‌ provided in the paper‌ "A Refined Mean Field‌​‌ Approximation" by N. Gast​​ and B. Van Houdt,​​​‌ SIGMETRICS 2018, and a‌ framework to compute heterogeneous‌​‌ mean field approximations as​​ proposed in "Mean Field​​​‌ and Refined Mean Field‌ Approximations for Heterogeneous Systems:‌​‌ It Works!" by N.​​ Gast and S. Allmeier,​​​‌ SIGMETRICS 2022.

• URL:
  https://­github.­com/­ngast/­rmf_tool‌
• Publications:
  hal-01622054, tel-02509756‌​‌, hal-03485044
• Contact:
  Nicolas​​ Gast

7.1.7 SCIP

• Name:​​​‌
  Solving Constraint Integer Programs‌
• Keywords:
  Linear optimization, Mathematical‌​‌ Optimization, Mixed Integer Programming​​
• Functional Description:
  SCIP is​​​‌ currently one of the‌ fastest non-commercial solvers for‌​‌ mixed integer programming (MIP)​​ and mixed integer nonlinear​​​‌ programming (MINLP). It is‌ also a framework for‌​‌ constraint integer programming and​​ branch-cut-and-price. It allows for​​​‌ total control of the‌ solution process and the‌​‌ access of detailed information​​ down to the guts​​​‌ of the solver.
• Release‌ Contributions:

  SCIP 10.0.0

  Features‌​‌ and Performance Improvements

  Exact​​ Solving:

  - added numerically​​​‌ exact solving mode for‌ mixed-integer linear programs to‌​‌ the core framework including​​ certification of branch-and-bound phase,​​​‌ - core extensions (new‌ wrapper struct SCIP_RATIONAL for‌​‌ rational arithmetic currently based​​ on Boost, GMP, and​​​‌ MPFR, new data structure‌ SCIP_LPEXACT for handling rational‌​‌ LP relaxation and computing​​ safe dual bounds, new​​​‌ interfaces to exact LP‌ solvers SoPlex and QSopt_ex,‌​‌ safe dualproof version of​​ conflict analysis, new data​​​‌ structure SCIP_CERTIFICATE for certificate‌ printing/proof logging) - new‌​‌ plugins: new constraint handler​​ "exactlinear" for handling linear​​​‌ constraints with rational data,‌ new constraint handler "exactsol"‌​‌ to post-process and repair​​ solutions from floating-point heuristics,​​​‌ - plugins revised for‌ numerically exact solving mode:‌​‌ adjusted readers for MPS,​​ LP, CIP, OPB/WBO, and​​​‌ ZIMPL files, extended presolver‌ "milp" to perform rational‌​‌ presolving with PaPILO, adjusted​​​‌ constraint handler "integral" and​ default reliability pseudo-cost branching​‌ rule "relpscost", extended Gomory​​ cut separator to separate​​​‌ and certify numerically safe​ MIR cuts, adjusted all​‌ primal heuristics (except for​​ five dedicated MINLP heuristics),​​​‌ new interfaces to exact​ LP solvers SoPlex and​‌ QSopt_ex

  Symmetry Handling

  -​​ added more techniques to​​​‌ handle reflection symmetries, in​ particular, for orbitopes with​‌ column reflections and matrices​​ whose rows and columns​​​‌ can be permuted by​ a symmetry - Dejavu​‌ can be used to​​ compute symmetries, the source​​​‌ code is shipped with​ SCIP and incorporates sassy​‌ - implemented symmetry detection​​ callbacks for disjunction and​​​‌ superindicator constraint handlers -​ detailed information about applied​‌ symmetry handling techniques can​​ be printed to the​​​‌ terminal - improve memory​ usage by introducing different​‌ constraint handlers for full​​ orbitopes and packing/partitioning orbitopes​​​‌ - symmetry detection no​ longer treats implicit integer​‌ variables separately, but computes​​ symmetries based on the​​​‌ variable type inferred from​ variable bounds and implied​‌ integrality - extended the​​ statistics to also include​​​‌ information about the number​ of variables (per type)​‌ affected by symmetry -​​ implemented method to compute​​​‌ new permutations from a​ given list of symmetry​‌ group generators - cons_orbisack,​​ cons_orbitope_full, cons_orbitope_pp, and cons_symresack​​​‌ now try to replace​ the stored aggregated variables​‌ by active ones at​​ the end of presolving,​​​‌ this should reduce the​ size of copies of​‌ the presolved problem simplified​​ symmetry detection graphs in​​​‌ case all edges have​ the same color

  Presolve:​‌

  - distinguish implicit integrality​​ of variables into strong​​​‌ and weak type, depending​ on whether integrality is​‌ implied for all feasible​​ or only at least​​​‌ one optimal solution -​ added a new presolver​‌ "implint", which detects implied​​ integral variables by detecting​​​‌ (transposed) network submatrices in​ the problem, for now,​‌ this plugin is disabled​​ by default - added​​​‌ support for (transposed) network​ matrix detection allow multi-aggregation​‌ of unbounded slack variables,​​ which may enable more​​​‌ bound tightening due to​ a reduction in the​‌ number of unbounded variables​​ resolve all fixings in​​​‌ xor constraints also for​ an available integer variable​‌

• URL:
  https://­www.­scipopt.­org/
• Contact:
  Mathieu​​ Besancon
• Partners:
  TU Darmstadt,​​​‌ RWTH Aachen University, Friedrich-Alexander-Universität​ Erlangen-Nürnberg, Eindhoven University of​‌ Technology, University of Twente,​​ University of Bayreuth, Forschungscampus​​​‌ Modal

Non-Stationary Gradient Descent for‌ Optimal Auto-Scaling in Serverless‌​‌ Platforms

To efficiently manage​​ serverless computing platforms, a​​​‌ key aspect is the‌ auto-scaling of services, i.e.,‌​‌ the set of computational​​ resources allocated to a​​​‌ service adapts over time‌ as a function of‌​‌ the traffic demand. The​​ objective is to find​​​‌ a compromise between user-perceived‌ performance and energy consumption.‌​‌ In 3, we​​ consider the scale-per-request auto-scaling​​​‌ pattern and investigate how‌ many function instances (or‌​‌ servers) should be spawned​​ each time an unfortunate​​​‌ job arrives, i.e., a‌ job that finds all‌​‌ servers busy upon its​​ arrival. We address this​​​‌ problem by following a‌ stochastic optimization approach: we‌​‌ develop a stochastic gradient​​ descent scheme of the​​​‌ Kiefer-Wolfowitz type that applies‌ over a single run‌​‌ of the state evolution.​​ At each iteration, the​​​‌ proposed scheme computes an‌ estimate of the number‌​‌ of servers to spawn​​ each time an unfortunate​​​‌ job arrives to minimize‌ some cost function. Under‌​‌ natural assumptions, we show​​ that the sequence of​​​‌ estimates produced by our‌ scheme is asymptotically optimal‌​‌ almost surely. In addition,​​ we prove that its​​​‌ convergence rate is $\mathrm{O‌}\left(n^{-\mathrm{2‌‌}/3}\right)$ where​​ $n$ is the number​​​‌ of iterations.

From a‌ mathematical point of view,‌​‌ the stochastic optimization framework​​ induced by auto-scaling exhibits​​​‌ non-standard aspects that we‌ approach from a general‌​‌ point of view. We​​ consider the setting where​​​‌ a controller can only‌ get samples of the‌​‌ transient – rather than​​ stationary – behavior of​​​‌ the underlying stochastic system.‌ To handle this difficulty,‌​‌ we develop arguments that​​ exploit properties of the​​​‌ mixing time of the‌ underlying Markov chain. By‌​‌ means of numerical simulations,​​ we validate the proposed​​​‌ approach and quantify its‌ gain with respect to‌​‌ common existing scale-up rules.​​

Autoscaling in Serverless Platforms​​​‌ via Online Learning with‌ Convergence Guarantees

As the‌​‌ adoption of serverless computing​​ platforms continue to grow,​​​‌ designing autoscaling policies that‌ strike the right balance‌​‌ between energy efficiency and​​ user-perceived performance has become​​​‌ a central challenge. In‌ 35, we propose‌​‌ an online learning algorithm​​ with theoretical convergence guarantees​​​‌ that dynamically tunes control‌ parameters in a serverless‌​‌ autoscaling environment. The proposed​​ algorithm, grounded in stochastic​​​‌ gradient descent, learns online-during‌ the actual operation of‌​‌ the platform-the optimal values​​ of three key control​​​‌ parameters: (i) the target‌ stock size of prewarmed‌​‌ (idle) functions, (ii) the​​ threshold triggering provisioning actions,​​​‌ and (iii) the expiration‌ rate of idle resources.‌​‌ We prove that, under​​ Markovian dynamics, the algorithm​​​‌ converges to the parameter‌ set that minimizes a‌​‌ cost function capturing the​​ tradeoff between energy consumption​​​‌ and response latency. In‌ addition, we demonstrate that‌​‌ its structure naturally supports​​ parallelization, significantly accelerating convergence.​​​‌

Extensive numerical experiments show‌ that our method outperforms‌​‌ existing baselines, including recent​​ deep learning-based approaches, even​​​‌ under non-Markovian settings-highlighting both‌ its robustness and practical‌​‌ viability for next-generation serverless​​ infrastructures.

Energy-Optimal Scheduling with​​​‌ Variable Processing Speed: The‌ Role of Task Size‌​‌ Variability

In 2,​​​‌ we study the execution​ of a single task​‌ with an unknown size​​ on a server with​​​‌ variable processing speed. Our​ goal is to analyze​‌ structural properties of the​​ optimal energy consumption under​​​‌ the optimal speed profile​ that minimizes the expected​‌ energy consumption while meeting​​ a hard deadline constraint.​​​‌ Specifically, we investigate how​ the task size probability​‌ distribution impacts the overall​​ energy.

Under mild assumptions,​​​‌ our main result shows​ that the expected energy​‌ consumption induced by the​​ optimal speed profile preserves​​​‌ the convex increasing order​ with respect to the​‌ task size distribution. Then,​​ we leverage this property​​​‌ to derive simple bounds​ and conduct a worst-case​‌ analysis. In particular, we​​ derive a simple, general​​​‌ formula for the energy​ gap induced by the​‌ 'best' and 'worst' task​​ size distributions, expressed in​​​‌ terms of the support​ and expectation of the​‌ task size.

Time-Constrained Energy​​ Minimization for Online Execution​​​‌ of a Stochastic DAG​ Task

In 30,​‌ We study the problem​​ of energy-efficient online execution​​​‌ of a complex task​ on a server with​‌ variable processing speed. The​​ task consists of a​​​‌ set of stochastic elementary​ jobs structured as a​‌ Directed Acyclic Graph (DAG),​​ where each job's execution​​​‌ may reveal new information​ that influences future scheduling​‌ decisions. Our objective is​​ to determine an online​​​‌ speed control policy that​ minimizes the expected energy​‌ consumption while ensuring that​​ the task completes before​​​‌ a strict deadline. Leveraging​ tools from convex optimization,​‌ the optimality principle, and​​ backward induction, we derive​​​‌ a structural characterization of​ the optimal policy. We​‌ find that this is​​ linked to a set​​​‌ of second-order differential equations​ and exhibits a non-trivial​‌ form. Building on this​​ result, we develop a​​​‌ discretization-based algorithm that efficiently​ approximates the optimal policy.​‌ The proposed algorithm is​​ provably asymptotically exact in​​​‌ the discretization step and​ has computational complexity $\mathrm{O‌}\left(\right|A|+KN)‌$, where $\left|\mathrm{A}\right|$ and $K$ denote​‌ the number of edges​​ and vertices (i.e., jobs)​​​‌ in the underlying DAG,​ and $N$ is the​‌ discretization granularity. Our results​​ offer a principled and​​​‌ computationally efficient solution framework​ for online execution of​‌ structured stochastic workloads under​​ strict energy and timing​​​‌ constraints.

8.2.1 Experimental​​​‌ practices and Simulation

8.2.2‌ Mean Field

As a‌​‌ system of stochastically interacting​​ entities grows large, the​​​‌ size of its state-space‌ quickly explodes (curse of‌​‌ dimensionality) and both its​​ analysis and optimization become​​​‌ intractable. Fortunately, symmetries and‌ regularities can be exploited,‌​‌ in particular through the​​ so-called Mean Field approximation,​​​‌ which averages out the‌ state over degrees of‌​‌ freedom. This technique from​​ statistical physics is particularly​​​‌ effective when studying computer‌ systems and over the‌​‌ years, we have refined​​ such approximation, proposed extensions,​​​‌ and developed fine analysis‌ methods.

Logarithmic regret of​​​‌ exploration in average reward​ Markov decision processes

In​‌ average reward Markov decision​​ processes, state-of-the-art algorithms for​​​‌ regret minimization follow a​ well-established framework: They are​‌ model-based, optimistic and episodic.​​ First, they maintain a​​​‌ confidence region from which​ optimistic policies are computed​‌ using a well-known subroutine​​ called Extended Value Iteration​​​‌ (EVI). Second, these policies​ are used over time​‌ windows called episodes, each​​ ended by the Doubling​​​‌ Trick (DT) rule or​ a variant thereof. In​‌ 13, without modifying​​ EVI, we show that​​​‌ there is a significant​ advantage in replacing (DT)​‌ by another simple rule,​​ that we call the​​​‌ Vanishing Multiplicative (VM) rule.​ When managing episodes with​‌ (VM), the algorithm's regret​​ is, both in theory​​​‌ and in practice, as​ good if not better​‌ than with (DT), while​​ the one-shot behavior is​​​‌ greatly improved. More specifically,​ the management of bad​‌ episodes (when sub-optimal policies​​ are being used) is​​​‌ much better under (VM)​ than (DT) by making​‌ the regret of exploration​​ logarithmic rather than linear.​​​‌ These results are made​ possible by a new​‌ in-depth understanding of the​​ contrasting behaviors of confidence​​​‌ regions during good and​ bad episodes.

8.4.1 Langevin Sampling​​​‌ and Large Deviation Techniques​

8.4.2 Frank-Wolfe‌​‌

8.4.3 Mixed Integer‌​‌ Optimization

8.4.4 Application to​​​‌ various domains

8.4.5​‌ Optimization for Machine Learning​​

Many learning algorithms operate​​​‌ in centralized way, which​ raises many practical issues​‌ in terms of scalability,​​ privacy, hence a high​​ interest for designing efficient​​​‌ distributed and federated machine‌ learning algorithms. Furthermore generally,‌​‌ the optimization space is​​ also quite particular, which​​​‌ calls for specific regularization‌ techniques and optimization algorithms.‌​‌

8.5.1​​ Learning in stochastic games​​​‌

8.5.2 Efficient​​​‌ Learning in Complex Landscape‌ Games

8.5.3 Uncertainty and​​ Robustness

Random​​​‌ Matrix and Tensor Models​ for Large Data Processing​‌

The exponential growth in​​ computing power has enabled​​​‌ the widespread deployment of​ machine learning, which has​‌ in turn given rise​​ to new challenges in​​​‌ data processing. The sheer​ volume of data now​‌ being generated means that​​ the standard statistical assumption​​​‌ of a number of​ samples far greater than​‌ their dimension is no​​ longer tenable. In the​​​‌ paradigm of the Big​ Data era, datasets are​‌ typically of very large​​ dimension and may also​​​‌ comprise several modes, indicating​ a variety of sources,​‌ modalities, domains, and so​​ on. Furthermore, the advancement​​​‌ of technologies required to​ develop models capable of​‌ processing vast quantities of​​ data results in significant​​​‌ environmental and human costs.​ In light of these​‌ concerns, it is imperative​​ to promote a more​​​‌ clever and prudent use​ of our resources.Random matrix​‌ theory provides powerful tools​​ to precisely study the​​​‌ statistical and computational limitations​ associated with the processing​‌ of large and multidimensional​​ data. Through this lens,​​ the thesis of Hugo​​​‌ Lebeau 29 examines several‌ learning approaches to identify‌​‌ the relevant parameters influencing​​ the success of a​​​‌ task and thereby facilitate‌ an informed use.

Firstly,‌​‌ we examine an extension​​ of spectral clustering to​​​‌ data streams. This approach‌ enables the clustering of‌​‌ a potentially very large​​ dataset with a controlled​​​‌ and limited memory usage.‌ Our findings demonstrate that,‌​‌ with an astute management​​ of the available memory,​​​‌ it is possible to‌ achieve performance levels comparable‌​‌ to those obtained without​​ resource constraints.We then turn​​​‌ our attention to the‌ computational limits to tensor‌​‌ estimation, with a particular​​ focus on low-rank approximation.​​​‌ The study the reconstruction‌ performance of the truncated‌​‌ MLSVD (which generalizes the​​ concept of truncated SVD​​​‌ to tensors) as well‌ as the HOOI algorithm‌​‌ precisely describes the conditions​​ required for the reconstruction​​​‌ of a noisy signal‌ in multidimensional data. Additionally,‌​‌ we utilize a similar​​ approach to investigate the​​​‌ multi-view clustering problem from‌ the perspective of a‌​‌ rank-one tensor approximation. Our​​ findings shed light on​​​‌ and precisely quantifies the‌ pivotal role of view‌​‌ informativeness in the quality​​ of the estimation.Lastly, we​​​‌ investigate the statistical limits‌ to tensor estimation by‌​‌ introducing a matrix model​​ associated to the maximum​​​‌ likelihood problem. This approach‌ allows us to characterize‌​‌ the reconstruction performance of​​ the corresponding estimator through​​​‌ a spectral analysis that‌ can be performed with‌​‌ the standard tools of​​ random matrix theory. To​​​‌ illustrate this method, we‌ examine the best rank-one‌​‌ approximation of a tensor,​​ where a given proportion​​​‌ of entries is randomly‌ removed to reduce its‌​‌ memory cost. This allows​​ us to quantify the​​​‌ impact of such a‌ procedure on the quality‌​‌ of the resulting estimate.​​ Finally, we propose to​​​‌ extend the presented method‌ to a more general‌​‌ tensor estimation framework, which​​ reveals attractive new challenges​​​‌ for the study of‌ large random tensors.

A‌​‌ Random Matrix Approach to​​ Low-Multilinear-Rank Tensor Approximation

9​​​‌ presents a comprehensive understanding‌ of the estimation of‌​‌ a planted low-rank signal​​ from a general spiked​​​‌ tensor model near the‌ computational threshold. Relying on‌​‌ standard tools from the​​ theory of large random​​​‌ matrices, we characterize the‌ large-dimensional spectral behavior of‌​‌ the unfoldings of the​​ data tensor and exhibit​​​‌ relevant signal-to-noise ratios governing‌ the detectability of the‌​‌ principal directions of the​​ signal. These results allow​​​‌ to accurately predict the‌ reconstruction performance of truncated‌​‌ multilinear SVD (MLSVD) in​​ the non-trivial regime. This​​​‌ is particularly important since‌ it serves as an‌​‌ initialization of the higher-order​​ orthogonal iteration (HOOI) scheme,​​​‌ whose convergence to the‌ best low-multilinear-rank approximation depends‌​‌ entirely on its initialization.​​ We give a sufficient​​​‌ condition for the convergence‌ of HOOI and show‌​‌ that the number of​​ iterations before convergence tends​​​‌ to 1 in the‌ large-dimensional limit.

Performance of‌​‌ Rank-One Tensor Approximation on​​ Incomplete Data

In 26​​​‌, we are interested‌ in the estimation of‌​‌ a rank-one tensor signal​​ when only a portion​​​‌ $ϵ$ of its noisy‌ observation is available. We‌​‌ show that the study​​​‌ of this problem can​ be reduced to that​‌ of a random matrix​​ model whose spectral analysis​​​‌ gives access to the​ reconstruction performance. These results​‌ shed light on and​​ specify the loss of​​​‌ performance induced by an​ artificial reduction of the​‌ memory cost of a​​ tensor via the deletion​​​‌ of a random part​ of its entries.

Prophet Inequalities:​​ Competing with the Top​​​‌ $\ell$ Items is Easy​

In 23, we​‌ explore a prophet inequality​​ problem, where the values​​​‌ of a sequence of​ items are drawn i.i.d.​‌ from some distribution, and​​ an online decision maker​​​‌ must select one item​ irrevocably. We establish that​‌ ${\text{CR}}_{\ell }$ the worst-case​​ competitive ratio between the​​​‌ expected optimal performance of​ an online decision maker​‌ compared to that of​​ a prophet who uses​​​‌ the average of the​ top $\ell$ items is​‌ exactly the solution to​​ an integral equation. This​​​‌ quantity ${\text{CR}}_{\ell }$ is​ larger than $1-‌e^{-\ell }$.​​ This implies that the​​​‌ bound converges exponentially fast​ to 1 as $\mathrm{\ell ‌}$ grows. In particular for​​ $\ell =2$,​​​‌ $C{R}_2\approx 0.966$,​‌ which is much closer​​ to 1 than the​​​‌ classical bound of $\mathrm{0}.745$ for $\mathrm{\ell ‌}=1$. Additionally,​​ we prove asymptotic lower​​​‌ bounds for the competitive​ ratio of a more​‌ general scenario, where the​​ decision maker is permitted​​​‌ to select $k$ items.​ This subsumes the $\mathrm{k‌}$ multi-unit i.i.d. prophet problem​​ and provides the current​​​‌ best asymptotic guarantees, as​ well as enables broader​‌ understanding in the more​​ general framework. Finally, we​​​‌ prove a tight asymptotic​ competitive ratio when only​‌ static threshold policies are​​ allowed.

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

10.2.1‌ Visits of international scientists‌​‌

10.2.2 Visits​​​‌ to international teams

10.3.1 Horizon Europe

Panayotis​​ Mertikopoulos participated in the​​​‌ submission of the QU-METRIC‌ proposal to the call‌​‌ HORIZON-EIC-2025-PATHFINDEROPEN. The proposal was​​ unsuccessful; unclear if there​​​‌ will be a follow-up‌ with the same consortium.‌​‌

ANR

• ANR REFINO​​ (JCJC 2020-2025)$☆$

  Refined​​​‌ Mean Field Optimization[250K€]‌

  REFINO is an ANR‌​‌ starting grant (JCJC) coordinated​​ by Nicolas Gast. The​​​‌ main objective on this‌ project is to provide‌​‌ an innovative framework for​​ optimal control of stochastic​​​‌ distributed agents. Restless bandit‌ allocation is one particular‌​‌ example where the control​​ that can be sent​​​‌ to each arm is‌ restricted to an on/off‌​‌ signal. The originality of​​ this framework is the​​​‌ use of refined mean‌ field approximation to develop‌​‌ control heuristics that are​​ asymptotically optimal as the​​​‌ number of arms goes‌ to infinity and that‌​‌ also have a better​​ performance than existing heuristics​​​‌ for a moderate number‌ of arms. As an‌​‌ example, we will use​​ this framework in the​​​‌ context of smart grids,‌ to develop control policies‌​‌ for distributed electric appliances.​​

• ANR FAIRPLAY (JCJC 2021-2025)​​​‌$☆$

  Fair algorithms via‌ game theory and sequential‌​‌ learning[245K€]

  FAIRPLAY is​​ an ANR starting grant​​​‌ (JCJC) coordinated by Patrick‌ Loiseau. Machine learning algorithms‌​‌ are increasingly used to​​ optimize decision making in​​​‌ various areas, but this‌ can result in unacceptable‌​‌ discrimination. The main objective​​ of this project is​​​‌ to propose an innovative‌ framework for the development‌​‌ of learning algorithms that​​ respect fairness constraints. While​​​‌ the literature mostly focuses‌ on idealized settings, the‌​‌ originality of this framework​​ and central focus of​​​‌ this project is the‌ use of game theory‌​‌ and sequential learning methods​​ to account for constraints​​​‌ that appear in practical‌ applications: strategic and decentralized‌​‌ aspects of the decisions​​​‌ and the data provided​ and absence of knowledge​‌ of certain parameters key​​ to the fairness definition.​​​‌

11.1.1​​​‌ Scientific events: selection

Panayotis​ Mertikopoulos has been senior​‌ area chair at NeurIPS​​ 2025 and area chair​​​‌ at ICML 2025.

11.1.2​​​‌ Journal

11.1.3 Invited talks​​​‌

• Arnaud Legrand was invited​ to give lectures and​‌ keynotes on Reproducible Research​​ and Open Science on​​ the following occasions:
  • Congrès​​​‌ de la ROADEF, Champ-sur-Marne‌ (Feb. 2025)
  • École thématique‌​‌ Science Ouverte pour les​​ SHS, Oléron (June​​​‌ 2025)
  • M1 students in‌ computer science at UGA‌​‌ (Dec. 2025): Reproducible Research​​ and Computer Science
  • 1st​​​‌ year Inria PhD students‌ (Nov. 2025)
• Bruno Gaujal‌​‌ and Nicolas Gast were​​ invited to give a​​​‌ tutorial on Stochastic bandits‌ at Performance 2025 (Amsterdam).‌​‌
• Bruno Gaujal was invited​​ to a roundtable on​​​‌ the future of performance‌ evaluation at Atelier Evaluation‌​‌ des Performances (Toulouse).
• Panayotis​​ Mertikopoulos was invited to​​​‌ present his work at‌ the following events:
  • Invited‌​‌ conference talk: 75 years​​ of Nash equilibrium Oxford,​​​‌ UK
  • Invited workshop talk:‌ 3rd Paris Workshop on‌​‌ Game Theory and Language​​
  • Tutorial talk at the​​​‌ 7th Annual Conference on‌ Learning for Dynamics &‌​‌ Control
  • Invited seminar talks​​ at: LSE, MCGT

11.1.4​​​‌ Leadership within the scientific‌ community

• Mathieu Besancon is‌​‌ a member of the​​ MOBIDEC PEPR in the​​​‌ ACME project.
• Panayotis Mertikopoulos‌ is the scientific coordinator‌​‌ of the PEPR IA​​ projet ciblé (acceleration grant)​​​‌ FOUNDRY: Foundations of robustness‌ and reliability in artificial‌​‌ intelligence. The project​​ has a total budget​​​‌ of 5M€ and involves‌ five research teams across‌​‌ France (POLARIS in Grenoble,​​ the Inria teams SCOOL​​​‌ and FAIRPLAY, Dauphine and‌ IMT in Paris, and‌​‌ ENS Lyon).
• Jean-Marc Vincent​​ is a member of​​​‌ the scientific committee of‌ the CIST.
• Jean-Marc Vincent‌​‌ is vice-head of the​​ SIF, adjunct on teaching.​​​‌ In this context he‌ is in charge of‌​‌ the organization of the​​ annual meeting on education​​​‌ in computer science.

11.1.5‌ Research administration

• Mathieu Besancon‌​‌ has been a member​​ of the hiring committee​​​‌ for an Assistant Professor‌ at Univ. Clermont Auvergne.‌​‌
• Nicolas Gast is vice-head​​ of the Labex EnergyAlps​​​‌ that federates the community‌ working on electrical energy‌​‌ in Grenoble.
• Nicolas Gast​​ is vice-head of the​​​‌ école doctorale MSTII (the‌ doctoral school managing PhD‌​‌ students in computer science​​ and mathemathics at Univ.​​​‌ Grenoble Alpes).
• Bruno Gaujal‌ was president of the‌​‌ hiring community for and​​ Assistant Professor at Grenoble​​​‌ INP-GI.
• Arnaud Legrand is‌ a member of the‌​‌ Section 6 of the​​ CoNRS.
• Arnaud Legrand is​​​‌ head of the SRCPR‌ axis of the LIG‌​‌ and a member of​​ LIG bureau. He has​​​‌ been particularly involved in‌ the HCERES evaluation process‌​‌ of the LIG.
• Arnaud​​ Legrand h as been​​​‌ a member of the‌ HCERES evaluation committee of‌​‌ the IRIT.
• Arnaud Legrand​​ is a member of​​​‌ Comité Scientique of the‌ Inria Grenoble.
• Florence Perronin‌​‌ is a member of​​ the QVT team of​​​‌ the LIG.
• Jean-Marc Vincent‌ was a member of‌​‌ the jury for Mcf​​ hiring in Univ. Nancy.​​​‌

11.2.1​‌ Supervision

Bruno Gaujal is​​ member of the CSI​​​‌ of Vittorio Puricelli (Toulouse)​ and Alessia Rigonat (Paris).​‌

11.2.2 Juries

• Nicolas Gast​​ was president of the​​​‌ PhD jury of Bianca​ Marin Moreno (Univ. Grenoble​‌ Alpes) entitled Apprentissage par​​ renforcement convexe en ligne​​​‌ et applications aux problèmes​ de gestion de l'énergie​‌.
• Nicolas Gast was​​ reviewer for the PhD​​​‌ of Thomas Le Corre​ (ENS PSL) entitled Distributed​‌ control of flexible loads​​ in power grids.​​​‌
• Bruno Gaujal was a​ member of the jury​‌ of the PhD defense​​ of Lucas Weber (Paris)​​​‌ entitled Exploiting Partial System​ Knowledge in Reinforcement Learning​‌ for Admission Control and​​ Electricity Storage Optimization,​​​‌ president of the jury​ of the PhD of​‌ Andrea Fox (Avigon) entitled​​ Reinforcement Learning for Resource​​ Allocation in Edge/Fog Systems​​​‌ , and reviewer for‌ the Phd of Maria‌​‌ Cherifa (Saclay) entitled Dynamics​​ and learning in some​​​‌ onlineallocation problems and of‌ the PhD of Chiara‌​‌ Mignacco (Saclay) entitled A​​ mathematical study of policy​​​‌ orchestration for reinforcement learning‌.
• Bruno Gaujal was‌​‌ president of the HDR​​ jury of Pierre Gaillard​​​‌ (Grenoble)
• Arnaud Legrand was‌ a reviewer for the‌​‌ PhD of Léo Cosseron​​ (ÉNS Rennes): Time-accurate network​​​‌ simulation interconnecting virtual machines‌ with hardware virtualization towards‌​‌ stealth analysis.
• Arnaud​​ Legrand was president of​​​‌ the PhD jury of‌ Eduardo Tomasi Ribeiro (Univ.‌​‌ Grenoble Alpes): Single address​​ space for 128-bit massively​​​‌ parallel computers.

11.3.1 Specific official‌​‌ responsibilities in science outreach​​ structures

• Jean-Marc Vincent is​​​‌ the head of the‌ evaluation committee for education‌​‌ in AI of the​​ MIAI chaire.

International journals‌​‌

• 1 articleS.Sebastian​​ Allmeier and N.Nicolas​​​‌ Gast. Accuracy of‌ the Graphon Mean Field‌​‌ Approximation for Interacting Particle​​ Systems.Stochastic Systems​​​‌154December 2025‌, 273-290HALDOI‌​‌back to text
• 2​​ articleJ.Jonatha Anselmi​​​‌ and B.Bruno Gaujal‌. Energy-Optimal Scheduling with‌​‌ Variable Processing Speed: The​​ Role of Task Size​​​‌ Variability.Operations Research‌ Letters64December 2025‌​‌, 107379HALDOI​​back to text
• 3​​​‌ articleJ.Jonatha Anselmi‌, B.Bruno Gaujal‌​‌ and L.-S.Louis-Sébastien Rebuffi​​. Non-Stationary Gradient Descent​​​‌ for Optimal Auto-Scaling in‌ Serverless Platforms.IEEE/ACM‌​‌ Transactions on Networking2025​​, 1-14HALDOI​​​‌back to text
• 4‌ articleM.Mathieu Besançon‌​‌, S.Sébastien Designolle​​, J.Jannis Halbey​​​‌, D.Deborah Hendrych‌, D.Dominik Kuzinowicz‌​‌, S.Sebastian Pokutta​​, H.Hannah Troppens​​​‌, D.Daniel Viladrich‌ Herrmannsdoerfer and E.Elias‌​‌ Wirth. Improved algorithms​​ and novel applications of​​​‌ the FrankWolfe.jl library.‌ACM Transactions on Mathematical‌​‌ SoftwareSeptember 2025,​​ 1-31HALDOIback​​​‌ to text
• 5 article‌M.Martin Bichler,‌​‌ D.Davide Legacci,​​ M.Matthias Oberlechner,​​​‌ P.Panayotis Mertikopoulos and‌ B.Bary Pradelski.‌​‌ Characterizing the Convergence of​​ Game Dynamics via Potentialness​​​‌.Transactions on Machine‌ Learning Research JournalMarch‌​‌ 2025, 1-22HAL​​back to text
• 6​​​‌ articleH.Henri Casanova‌, A.Arnaud Giersch‌​‌, A.Arnaud Legrand​​, M.Martin Quinson​​​‌ and F.Frédéric Suter‌. Lowering entry barriers‌​‌ to developing custom simulators​​ of distributed applications and​​​‌ platforms with SimGrid.‌Parallel Computing123March‌​‌ 2025, 103125HAL​​DOIback to text​​​‌
• 7 articleF.Fryderyk‌ Falniowski and P.Panayotis‌​‌ Mertikopoulos. On the​​ discrete-time origins of the​​​‌ replicator dynamics: From convergence‌ to instability and chaos‌​‌.International Journal of​​ Game Theory547​​​‌February 2025HALDOI‌back to text
• 8‌​‌ articleD.Deborah Hendrych​​, H.Hannah Troppens​​​‌, M.Mathieu Besançon‌ and S.Sebastian Pokutta‌​‌. Convex mixed-integer optimization​​ with Frank–Wolfe methods.​​​‌Mathematical Programming ComputationJune‌ 2025, 1-49HAL‌​‌DOIback to text​​​‌
• 9 articleH.Hugo​ Lebeau, F.Florent​‌ Chatelain and R.Romain​​ Couillet. A Random​​​‌ Matrix Approach to Low-Multilinear-Rank​ Tensor Approximation.Journal​‌ of Machine Learning Research​​2672025,​​​‌ 1-64HALback to​ text
• 10 articleF.​‌Francisca Vasconcelos, E.-V.​​Emmanouil-Vasileios Vlatakis-Gkaragkounis, P.​​​‌Panayotis Mertikopoulos, G.​Georgios Piliouras and M.​‌ I.Michael I Jordan​​. A Quadratic Speedup​​​‌ in Finding Nash Equilibria​ of Quantum Zero-Sum Games​‌.Quantum9May​​ 2025, 1737HAL​​​‌DOIback to text​

International peer-reviewed conferences

• 11​‌ inproceedingsW.Waïss Azizian​​, F.Franck Iutzeler​​​‌, J.Jérôme Malick​ and P.Panayotis Mertikopoulos​‌. The global convergence​​ of stochastic gradient descent​​​‌ in non-convex landscapes: Sharp​ estimates via large deviations​‌.Proceedings of the​​ 42nd International Conference on​​​‌ Machine LearningICML 2025​ - 42nd International Conference​‌ on Machine LearningVancouver,​​ Canada2025, 1-67​​​‌HALback to text​
• 12 inproceedingsM.Mathieu​‌ Besançon. Efficient Sparse​​ Flow Decomposition Methods for​​​‌ RNA Multi-Assembly.ECAI​ 2025 - 28th European​‌ Conference on Artificial Intelligence​​ECAI 2025 - 28th​​​‌ European Conference on Artificial​ IntelligenceBologna, Italy2025​‌, 1-8HALback​​ to text
• 13 inproceedings​​​‌V.Victor Boone and​ B.Bruno Gaujal.​‌ Logarithmic regret of exploration​​ in average reward Markov​​​‌ decision processes.Proceedings​ of Thirty Eighth Conference​‌ on Learning TheoryProceedings​​ of Thirty Eighth Conference​​​‌ on Learning Theory291​291Lyon, France2025​‌, 454--533HALback​​ to text
• 14 inproceedings​​​‌P.-L.Pierre-Louis Cauvin,​ D.Davide Legacci and​‌ P.Panayotis Mertikopoulos.​​ The impact of uncertainty​​​‌ on regularized learning in​ games.Proceedings of​‌ the 42nd International Conference​​ on Machine LearningICML​​​‌ 2025 - 42nd International​ Conference on Machine Learning​‌Vancouver, Canada2025,​​ 1-50HALback to​​​‌ text
• 15 inproceedingsN.​Nicolas Gast and D.​‌Dheeraj Narasimha. Model​​ Predictive Control is Almost​​​‌ Optimal for Restless Bandit​.COLT 2025 -​‌ 38th Annual Conference on​​ Learning TheoryLyon, France​​​‌2025, 1-35HAL​DOIback to text​‌
• 16 inproceedingsJ.Jannis​​ Halbey, S.Seta​​​‌ Rakotomandimby, M.Mathieu​ Besançon, S.Sébastien​‌ Designolle and S.Sebastian​​ Pokutta. Efficient Quadratic​​​‌ Corrections for Frank-Wolfe Algorithms​.NeurIPSSan Diego,​‌ United StatesDecember 2025​​HALback to text​​​‌
• 17 inproceedingsD.Deborah​ Hendrych, S.Sebastian​‌ Pokutta, M.Mathieu​​ Besançon and D.David​​​‌ Martínez-Rubio. Secant Line​ Search for Frank-Wolfe Algorithms​‌.Proceedings of Machine​​ Learning ResearchICML 2025​​​‌ - Forty-Second International Conference​ on Machine LearningVancouver​‌ (BC), CanadaJuly 2025​​, 1-22HALback​​​‌ to text
• 18 inproceedings​A.Andreas Kontogiannis,​‌ V.Vasilis Pollatos,​​ G.Gabriele Farina,​​​‌ P.Panayotis Mertikopoulos and​ I.Ioannis Panageas.​‌ Efficient kernelized learning in​​ polyhedral games beyond full-information:​​​‌ From Colonel Blotto to​ congestion games.NeurIPS​‌ 2025 - Proceedings of​​ the 39th International Conference​​​‌ on Neural Information Processing​ Systems, 2025San Diego,​‌ United StatesSeptember 2025​​HALback to text​​
• 19 inproceedingsK.Kyriakos​​​‌ Lotidis, P.Panayotis‌ Mertikopoulos, N.Nicholas‌​‌ Bambos and J.Jose​​ Blanchet. Invariance and​​​‌ concentration properties of gradient-based‌ learning in games.‌​‌CDC 2025 - 64th​​ IEEE Annual Conference on​​​‌ Decision and ControlRio‌ de Janeiro, BrazilIEEE‌​‌2025, 1-8HAL​​back to text
• 20​​​‌ inproceedingsK.Kyriakos Lotidis‌, P.Panayotis Mertikopoulos‌​‌, N.Nicholas Bambos​​ and J.Jose Blanchet​​​‌. Multi-agent learning under‌ uncertainty: Recurrence vs. concentration‌​‌.NeurIPS 2025 -​​ Proceedings of the 39th​​​‌ International Conference on Neural‌ Information Processing Systems, 2025‌​‌San Diego, United States​​December 2025HALback​​​‌ to textback to‌ text
• 21 inproceedingsK.‌​‌Kyriakos Lotidis, P.​​Panayotis Mertikopoulos, N.​​​‌Nicholas Bambos and J.‌Jose Blanchet. Robust‌​‌ equilibria in continuous games:​​ From strategic to dynamic​​​‌ robustness.NeurIPS 2025‌ - Proceedings of the‌​‌ 39th International Conference on​​ Neural Information Processing Systems,​​​‌ 2025San Diego, United‌ States2025, 1-33‌​‌HALback to text​​
• 22 inproceedingsI.Iosif​​​‌ Lytras and P.Panayotis‌ Mertikopoulos. Tamed Langevin‌​‌ sampling under weaker conditions​​.AISTATS 2025 -​​​‌ 28th International Conference on‌ Artificial Intelligence and Statistics‌​‌Mai Khao, Thailand2025​​, 1-32HALback​​​‌ to text
• 23 inproceedings‌M.Mathieu Molina,‌​‌ N.Nicolas Gast,​​ P.Patrick Loiseau and​​​‌ V.Vianney Perchet.‌ Prophet Inequalities: Competing with‌​‌ the Top $$ Items is​​ Easy.SODA 2025​​​‌ - ACM-SIAM Symposium on‌ Discrete AlgorithmsNew Orleans,‌​‌ United StatesACM2025​​, 1-45HALback​​​‌ to text
• 24 inproceedings‌H.Hannah Troppens,‌​‌ M.Mathieu Besançon,​​ S. E.St. Elmo​​​‌ Wilken and S.Sebastian‌ Pokutta. Mixed-Integer Optimization‌​‌ for Loopless Flux Distributions​​ in Metabolic Networks.​​​‌23rd International Symposium on‌ Experimental Algorithms (SEA 2025)‌​‌SEA 2025 - 23rd​​ International Symposium on Experimental​​​‌ Algorithms338Venice, Italy‌Schloss Dagstuhl – Leibniz-Zentrum‌​‌ für Informatik2025HAL​​DOIback to text​​​‌
• 25 inproceedingsE.Elias‌ Wirth, M.Mathieu‌​‌ Besançon and S.Sebastian​​ Pokutta. The Pivoting​​​‌ Framework: Frank-Wolfe Algorithms with‌ Active Set Size Control‌​‌.Proceedings of Machine​​ Learning ResearchAISTATS 2025​​​‌ - 28th International Conference‌ on Artificial Intelligence and‌​‌ StatisticsMai Khao, Thailand​​May 2025, 1-24​​​‌HALback to text‌

National peer-reviewed Conferences

• 26‌​‌ inproceedingsH.Hugo Lebeau​​. Performance of Rank-One​​​‌ Tensor Approximation on Incomplete‌ Data.GRETSI 2025‌​‌ – XXXème Colloque Francophone​​ de Traitement du Signal​​​‌ et des ImagesStrasbourg,‌ France2025, 1-4‌​‌HALback to text​​

Conferences without proceedings

• 27​​​‌ inproceedingsM.Mathieu Besançon‌. Frank-Wolfe Algorithms: Sparsity‌​‌ Guarantees and an Application​​ to Robust Optimization.​​​‌ROADEF 2025 - 26ème‌ Congrès annuel de la‌​‌ Société française de recherche​​ opérationnelle et d'aide à​​​‌ la décisionChamps-Sur-Marne, France‌February 2025, 1-1‌​‌HALback to text​​

Doctoral dissertations and habilitation​​​‌ theses

• 28 thesisR.‌Romain Cravic. Learning‌​‌ in stochastic games.​​UGA - Université Grenoble​​​‌ AlpesNovember 2025HAL‌back to textback‌​‌ to text
• 29 thesis​​​‌H.Hugo Lebeau.​ Random Matrix and Tensor​‌ Models for Large Data​​ Processing.Université Grenoble​​​‌ Alpes [2020-....]March 2025​HALback to text​‌back to text

Reports​​ & preprints

• 30 misc​​​‌J.Jonatha ANSELMI,​ B.Bruno Gaujal and​‌ K.Karl Gottlieb.​​ Time-Constrained Energy Minimization for​​​‌ Online Execution of a​ Stochastic DAG Task.​‌2025HALback to​​ text
• 31 reportN.​​​‌Nathalie Appel, J.​Jérémie Bourdon, N.​‌Nicolas Bousquet, J.​​Johanne Cohen, A.​​​‌Antoine Genitrini, P.​Patricia Georgeon, Y.​‌Yves Grandvalet, K.​​Katia Jaffrès-Runser, A.​​​‌Arnaud Legrand, D.​Damian Markham, A.​‌Anca Muscholl, A.​​Anastasia Paparrizou, L.​​​‌Loïc Paulevé, M.​Michael Poss, M.​‌Maria Gradinariu Potop-Butucaru,​​ J.-F.Jean-Florent Raymond,​​​‌ R.Romain Rouvoy,​ Y.Yolande Sallent,​‌ P.Pierre Senellart,​​ T.Thomas Seiller,​​​‌ Y.-Q.Ye-Qiong Song,​ A.Alain Tchana and​‌ H.Hélène Waeselynck.​​ Section 06 Sciences de​​​‌ l’information : fondements de​ l’informatique, calculs, algorithmes, représentations,​‌ exploitations: Rapport de conjoncture​​ 2024.CNRS2025​​​‌, 1-20HAL
• 32​ miscH.Hélène Arvis​‌, O.Olivier Beaude​​, N.Nicolas Gast​​​‌, S.Stéphane Gaubert​ and B.Bruno Gaujal​‌. Integrating Aggregated Electric​​ Vehicle Flexibilities in Unit​​​‌ Commitment Models using Submodular​ Optimization.November 2025​‌HALback to text​​
• 33 miscM.Mathieu​​​‌ Bacou, D.David​ Beserra, E.Eugen​‌ Dedu, L.Loïc​​ Desgeorges, D.Didier​​​‌ Donsez, A.Alexandre​ Guitton, B.Baptiste​‌ Jonglez, A.Arnaud​​ Legrand, G.Georgios​​​‌ Papadopoulos, O.Olivier​ Richard, S.Samir​‌ Si-Mohammed, N.Nina​​ Tamdrari and F.Fabrice​​​‌ Theoleyre. Journée thématique​ du GDR RSD :​‌ pratiques expérimentales de la​​ communauté systèmes et réseaux​​​‌.January 2025HAL​back to text
• 34​‌ reportC.Christopher Hojny​​, M.Mathieu Besançon​​​‌, K.Ksenia Bestuzheva​, S.Sander Borst​‌, A.Antonia Chmiela​​, J.João Dionísio​​​‌, J.Johannes Ehls​, L.Leon Eifler​‌, M.Mohammed Ghannam​​, A.Ambros Gleixner​​​‌, A.Adrian Göß​, A.Alexander Hoen​‌, J.Jacob von​​ Holly-Ponientzietz, R.Rolf​​​‌ van der Hulst,​ D.Dominik Kamp,​‌ T.Thorsten Koch,​​ K.Kevin Kofler,​​​‌ J.Jurgen Lentz,​ M.Marco Lübbecke,​‌ S. J.Stephen J.​​ Maher, P. M.​​​‌Paul Matti Meinhold,​ G.Gioni Mexi,​‌ T.Til Mohr,​​ E.Erik Mühmer,​​​‌ K. K.Krunal Kishor​ Patel, M. E.​‌Marc E. Pfetsch,​​ S.Sebastian Pokutta,​​​‌ C. R.Chantal Reinartz​ Groba, F.Felipe​‌ Serrano, Y.Yuji​​ Shinano, M.Mark​​​‌ Turner, S.Stefan​ Vigerske, M.Matthias​‌ Walter, D.Dieter​​ Weninger and L.Liding​​​‌ Xu. The SCIP​ Optimization Suite 10.0.​‌Zuse Institute BerlinNovember​​ 2025HALback to​​​‌ text
• 35 miscA.​ W.Abednego Wamuhindo Kambale​‌, J.Jonatha ANSELMI​​, D.Danilo Ardagna​​ and B.Bruno Gaujal​​​‌. Autoscaling in Serverless‌ Platforms via Online Learning‌​‌ with Convergence Guarantees.​​2025HALback to​​​‌ text
• 36 miscI.‌Iosif Lytras and P.‌​‌Panayotis Mertikopoulos. Contractive​​ kinetic Langevin samplers beyond​​​‌ global Lipschitz continuity.‌September 2025HALback‌​‌ to text

Other scientific​​ publications

• 37 inproceedings D.​​​‌Dorian Baudry, E.‌Emmeran Johnson, S.‌​‌Simon Vary, C.​​Ciara Pike-Burke and P.​​​‌Patrick Rebeschini. Does‌ Stochastic Gradient really succeed‌​‌ for Bandits? NeurIPS 2025​​ - 39th Conference on​​​‌ Neural Information Processing Systems‌ San Diego (Californie -‌​‌ EU), United States December​​ 2025 HAL back to​​​‌ text back to text‌
• 38 inproceedingsX.Xiaoqi‌​‌ Liu, D.Dorian​​ Baudry, J.Julian​​​‌ Zimmert, P.Patrick‌ Rebeschini and A.Arya‌​‌ Akhavan. Non-stationary Bandit​​ Convex Optimization: A Comprehensive​​​‌ Study.NeurIPS 2025‌ - 39th Conference on‌​‌ Neural Information Processing Systems​​San Diego (Californie -​​​‌ EU), United StatesDecember‌ 2025HALback to‌​‌ text

Software

• 39 software​​H.Henri Casanova,​​​‌ A.Augustin Degomme,‌ A.Arnaud Giersch,‌​‌ A.Arnaud Legrand,​​ M.Martin Quinson and​​​‌ F.Frédéric Suter.‌ SimGrid.4.0April‌​‌ 2025 lic: GNU Lesser​​ General Public License v2.1​​​‌ or later.HAL‌Software HeritageVCSback‌​‌ to text