Agenda depuis 2026
Seminaire : CANA seminar: Martin Pujol
16/11/2026 à 14h00
Title: Quantum algorithm for Ideals Fluids via Hamiltonian simulationRoom: 04.05 TPR2
Abstract: Quantum computers offer theoretical speedups for selected computational problems, but designing quantum algorithms for nonlinear partial differential equations remains challenging. In this work, we develop a quantum algorithm for the two-dimensional incompressible Euler equations by exploiting their Hamiltonian structure together with suitable discretizations. A Fourier–Galerkin truncation preserves the main conservative properties of the equations and can, in the best-case regime, converge exponentially with the truncation order, as confirmed numerically. The resulting nonlinear ODE system is embedded into a Liouville formulation and discretized into a Hermitian evolution problem suitable for quantum Hamiltonian simulation. The structure of the equations and of the discretization allows the elementary evolution operators to be implemented exactly with quantum circuits using resources polynomial in the number of degrees of freedom. We introduce a Trotter-based quantum algorithm, provide explicit quantum circuits, and analyze its approximation errors and computational complexity. Observables of the original equations are reconstructed using quantum amplitude estimation. Numerical experiments and classical emulations support the theoretical analysis and illustrate the convergence of the proposed quantum algorithm.
Seminaire : CANA seminar: Colin Geniet
03/11/2026 à 14h00
Title: Reducing CMSO to unbreakable graphs cannot be computableRoom: 04.05 TPR2
Abstract: In groundbreaking work, Lokshtanov, Ramanujan, Saurabh, and Zehavi [ICALP ‘18] proved that for any fixed CMSO formula φ, testing φ on arbitrary graphs can be reduced to testing φ on highly-connected graphs, in the sense of unbreakability. We prove that the parameters in this theorem are unfortunately inherently uncomputable. I will use this as an excuse to present the LRSZ result, the definitions, statement, and some intuition on the proof techniques, and then finish with our short uncomputability proof. This is joint work with Roohani Sharma.
Seminaire : Séminaire R2I invité : Manuel Castillo-Cara
27/10/2026 à 13h30
Title: Improving Deep Learning on Tabular Data by Exploiting Synthetic Images Abstract: Despite the breakthroughs of Deep Learning in computer vision and natural language processing, tabular data (which constitutes more than 80% of real-world corporate and scientific datasets) remains dominated by decision-tree-based algorithms due to the lack of inherent spatial relationships among variables. This keynote address introduces a paradigm shift based on converting tabular observations into two-dimensional synthetic images to unlock the full potential of vision architectures. Throughout the session, the TINTOlib ecosystem will be presented, analyzing the principal spatial generation methods in the literature (supervised and unsupervised, as well as parametric approaches such as TINTO, REFINED, and IGTD). Furthermore, the presentation will address advanced modeling through Convolutional Neural Networks (CNN), Vision Transformers (ViT), and prominently, Hybrid Neural Networks (HyNN) that fuse vision and tabular branches (CNN+MLP and ViT+MLP). Drawing on real-world use cases (e.g. indoor localization using Massive MIMO technology) and recent benchmarks, it will demonstrate how these hybrid architectures not only match or outperform classical baselines, but also act as learning stabilizers and regularizers. Finally, the latest developments in Explainable AI (XAI) tailored to this domain (such as PermGrad, Grad-CAM, and hybrid CNN–KAN models) will be examined, providing the means to audit and interpret the visual patterns extracted by models from transformed tabular data.Seminaire : CANA seminar: Sébastien Tavenas
21/10/2026 à 10h30
Title: Overview on the recent results on arithmetic circuits of constant depthRoom: 04.05 TPR2
Abstract: In recent years, there has been a strong rise of interest in constant-depth arithmetic circuits. To evaluate a polynomial on small inputs, it seems natural to perform the calculation using a sequence of arithmetic operations (addition, multiplication, subtraction, division, …). We can thus study computational models that are restricted to these operations. An arithmetic circuit is the equivalent of a Boolean circuit, but where OR, AND, NEG gates are replaced by +, *, /. One might then ask which polynomials can be computed using arithmetic circuits of polynomial size and constant depth. With Nutan and Srikanth, we had shown that it was not possible to compute the determinant of a matrix. But shortly thereafter, Andrews and Wigderson showed that it was still possible to compute a number of operations, such as the resultant or the GCD of two polynomials. During the talk, we will therefore review recent results concerning circuits of constant depth (lower bounds, upper bounds, closure).
Seminaire : Colloquium LIS - Samson Abramski
13/10/2026 à 09h00
La première édition du Colloquium du LIS aura lieu le mardi 13 octobre à partir de 9h à l'îlot Bernard Dubois (en face du campus Saint-Charles, https://feg.univ-amu.fr/fr/acces-campus#tab-3685), dans la salle de conférences au 3ème étage.
Nous aurons le plaisir d'accueillir Samson Abramsky, professeur émérite à l'Université d'Oxford (https://www.cs.ox.ac.uk/people/samson.abramsky/).
L'exposé sera accessible à l'ensemble de la communauté du LIS et mettra l'accent sur une vue d'ensemble autour de trois grands thèmes :
- Interaction (illustré par ses travaux sur la sémantique des jeux / game semantics)
- Compositionnalité (illustré par la mécanique quantique catégorique / Categorical Quantum Mechanics)
- Local vs Global (illustré par la contextualité / contextuality)
Une session posters se tiendra en parallèle de l'accueil café dès 9h (salle 15).
Programme :
- 09h00 : Accueil café (hall d'accueil) et session posters (salle 15)
- 10h00 : Exposé de Samson Abramsky
- 12h00 : Buffet
Au plaisir de vous y retrouver nombreux !
Lyes Benyoucef, Adrian Chifu, Basile Couëtoux
Seminaire : Seminaire IACD invité - Olivier Colliot (DR CNRS, ICM)
12/10/2026 à 14h00
Titre du séminaire : Caractérisation des données, reproductibilité, validation : l'IA comme une science expérimentale Date/Lieu: 12-10-2026, 14h, salle de commissions, campus St Jerome. Résumé : De nombreuses conclusions en IA reposent sur des arguments de nature expérimentale. Toutefois, les enjeux associés à cette nature expérimentale ne sont pas toujours suffisamment pris en compte. Ce thème sera abordé de trois points de vue : la reproductibilité, la caractérisation des données et la validation. Concernant la reproductibilité, j'introduirai un cadre conceptuel, distinguant différents types de reproductibilité. Je présenterai ensuite des travaux concernant la qualité des données dans le cadre d'entrepôts de données de santé où cette question est particulièrement prégnante. La troisième partie portera sur les méthodologies de validation, en particulier sur les liens entre caractéristiques d'une étude d'apprentissage automatique et incertitude sur les performances. Je terminerai par quelques résultats préliminaires dans le cas où les données sont caractérisées par une structure hiérarchique. La majorité des travaux seront illustrés dans le cadre de l'imagerie médicale. Mots-clés : qualité des données, entrepôts de données, évaluation de performances, reproductibilité, apprentissage automatique Bio : Olivier Colliot est Directeur de Recherche CNRS au sein de l'équipe ARAMIS de l'ICM (UMR 7225, CNRS, Sorbonne Université, Inserm, Inria). Depuis environ 20 ans, il travaille sur l'apprentissage automatique pour l'analyse d'images, avec pour application principale neuroimagerie. Ces dernières années, ses travaux ont débordé ce cadre applicatif spécifique. Ses thématiques de recherches actuelles porter sur la caractérisation des données, en particulier leur qualité, la reproductibilité, la validation et le benchmarking de méthodes d'apprentissage automatiques.Seminaire : CANA seminar: Nathan Claudet
29/09/2026 à 14h00
Title: Classifying quantum graph statesRoom: 04.05 TPR2
Abstract: Quantum computing is concerned with harnessing the peculiar properties of quantum mechanics, in order to perform information-processing tasks beyond the capabilities of classical computers. Graph states are a family of quantum states (the resources for quantum computers). Graph states exhibit complex forms of quantum entanglement but, unlike general quantum states, they are very easy to describe thanks to their one-to-one correspondence with mathematical graphs. I will discuss two important problems in graph-state theory: the classification of graph states according to their computational power as a quantum resource, and the classification of graph states according to their entanglement. In both instances, techniques from graph theory prove useful for tackling problems in quantum physics.
Seminaire : Séminaire COALA : Margaux SCHMIED (I3S Nice, équipe C&A)
02/07/2026 à 14h00
Lieu : Salle de réunion de l'équipe COALAUsing lower bound to improve the distance constraint on sequence variables
Abstract:
Insertion sequence variables have recently been introduced as a computational domain for modeling routing and sequencing problems in constraint programming. This paper investigates filtering for the (minimum) distance constraint over insertion sequence variables. This global constraint links a sequence to a distance variable based on a given distance matrix. So far, only a simple filtering algorithm has been proposed, which considers partial paths but ignores mandatory nodes. Our contribution introduces stronger lower bounds by integrating mandatory nodes. These bounds further enable the derivation of additional filtering rules for node insertions.