INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Sliding–Mode–Based Robust Interval Predictive Control for a Class of Uncertain and Constrained Systems
No full textInternational audienc
Processing Unstructured Meshes in Multithreaded Environments with the Help of Hilbert Renumbering and Dynamic Scheduling
No full textInternational audienceA software that deals with unstructured meshes is always tedious to parallelize because of its frequent indirect memory writes. These fine grain memory races are usually dealt with costly memory duplication and scatter-gather methods which require strong modifications in both algorithm and data structures. With the help of Hilbert renumbering and clever block dynamic scheduling, concurrent memory writes problems can be addressed at a very low compute and memory overhead. Furthermore, such a scheme always provides a very quick and easy parallelization of existing serial codes. It is, however, limited to shared memory machines but recent development in this area made tens of cores and terabytes of memory affordable, thus pushing the limit to billions of elements
Linear independence properties of the signature components of time-augmented stochastic processes
No full textThe addition of the running time as a component of a path before computing its signature is awidespread approach to ensure the one-to-one property between them [15] and leads to universalapproximation theorems [6, 7, 2]. However, this also leads to the linear dependence of the com-ponents of the terminal value of the signature of the time-augmented path. More precisely, for agiven natural number N , the signature components associated with words of length N – whichare computed with N iterated integrals – have the same linear span as the signature componentsassociated with words of length not greater than N . We generalize this result by exhibiting othersubfamilies of signature components with the same spanning properties. In particular we recoverthe result [8, Theorem 3.9] which states that the spanning of the iterated integrals with the lastintegrator different from the time variable is the same as the spanning of all iterated integrals.We check that this choice leads to the minimal computation time when the terms of the signatureare calculated using Chen’s relation in a backward way. The same optimal computation timeis symmetrically achieved in a forward way for the iterated integrals with the first integratordifferent from the time variable. Building on these results, we derive several results regardingthe linear independence of the signature components of a time-augmented stochastic process. Weshow that if the stochastic process we consider is solution to some SDE with additive Browniannoise then any subfamily of components proposed previously is linearly independent in L2. Wealso prove that the linear independence of these subfamilies of components is still true when weconsider the discretization of the sample paths of this stochastic process on a grid with a suffi-ciently small discretization time step. This property guarantees the relevance of such a selectionof features in numerical applications, particularly in regression tasks, when the true underlyingsignal is observed on a discrete time grid
Group-robust Machine Unlearning
No full textInternational audienceMachine unlearning is an emerging paradigm to remove the influence of specific training data (i.e., the forget set) from a model while preserving its knowledge of the rest of the data (i.e., the retain set). Previous approaches assume the forget data to be uniformly distributed from all training datapoints. However, if the data to unlearn is dominant in one group, we empirically show that performance for this group degrades, leading to fairness issues. This work tackles the overlooked problem of non-uniformly distributed forget sets, which we call group-robust machine unlearning, by presenting a simple, effective strategy that mitigates the performance loss in dominant groups via sample distribution reweighting. Moreover, we present MIU (Mutual Information-aware Machine Unlearning), the first approach for group robustness in approximate machine unlearning. MIU minimizes the mutual information between model features and group information, achieving unlearning while reducing performance degradation in the dominant group of the forget set. Additionally, MIU exploits sample distribution reweighting and mutual information calibration with the original model to preserve group robustness. We conduct experiments on three datasets and show that MIU outperforms standard methods, achieving unlearning without compromising model robustness. Source code available at https://github.com/tdemin16/group-robust_machine_unlearning
How important are inter-dataset interactions for large scale analysis of fMRI data: A multi-dimensional comparison
No full textInternational audienceAnalyzing multi-subject functional magnetic resonance imaging (fMRI) data requires methods that can jointly capture shared and individual patterns of brain activity across participants. Joint blind source separation (JBSS) techniques, such as independent vector analysis (IVA), provide a principled framework for this purpose by modeling dependencies across subjects while identifying distinct functional networks. Constrained IVA variants, including adaptive-reverse cIVA-G (ar-cIVA-G) and threshold-free cIVA-G (tf-cIVA-G), further enhance interpretability through the use of reference templates and inter-subject correlation constraints. Alternatively, regression-based methods like IVA-G regression (regIVA-G) and reference-guided component analysis (RGCA) process subjects individually, aligning their components to references with improved computational efficiency. Despite their potential, systematic evaluations of reference-based JBSS approaches for fMRI analysis remain limited. In this work, we present a comparative study of these 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 IVA and regression-based methods effectively extract meaningful biomarkers while the latter achieve comparable performance at substantially reduced computational cost
Efficient And Scalable Branch-and-Bound Algorithm for Exact Qubit Allocation
No full textInternational audienceQubit allocation is a central step in adapting abstract quantum circuits to noisy intermediate-scale quantum devices, yet exact approaches for solving it face severe scalability limitations. In this work, we revisit the formulation of qubit allocation as a permutation-based quadratic assignment problem and develop a branch-and-bound algorithm for its exact resolution. We first establish a refined sequential implementation that achieves significantly faster runtimes than previous exact approaches on most problem instances, thereby setting a new state-of-the-art for this formulation. Building on this foundation, we extend the approach to a performance-aware parallel implementation that exploits both intra-node and inter-node parallelism on High-Performance Computing (HPC) infrastructures. Our experimental evaluation demonstrates near-linear strong scaling at the intra-node level and substantial scalability in distributed settings across nodes. Leveraging these capabilities, we provide reference optimal solutions for challenging benchmark circuits of up to 26 qubits—significantly larger than previously reported instances. These results show that large-scale parallelization can effectively extend the reach of exact methods for qubit allocation, thereby advancing the integration of combinatorial optimization and HPC techniques in quantum computing
Coding computational laws: 20 recommendations for public administrations
No full textPublic administrations are steadily digitalizing all their procedures. In particular, computational laws – such as taxes and benefits – are increasingly implemented within computers, enabling scalable, automated computations. These computer implementations have four key specificities: they are critical software at the intersection between law and computer science, that will be updated regularly by legal changes, and have a long lifespan, counted in decades. Thus, great care should be taken to avoid any issue in these specific legal implementations. Building upon years of studying and coding computational laws, both in administrations and as new research products, we propose 20 recommendations to ease the development and maintenance of legal implementations. These recommendations aim at being understandable for lawyers and computer scientists alike
Beating Harmful Stereotypes Through Facts: RAG-based Counter-speech Generation
No full textInternational audienceCounter-speech generation is at the core of many expert activities, such as fact-checking and hate speech, to counter harmful content. Yet, existing work treats counter-speech generation as pure text generation task, mainly based on Large Language Models or NGO experts. These approaches show severe drawbacks due to the limited reliability and coherence in the generated countering text, and in scalability, respectively. To close this gap, we introduce a novel framework to model counter-speech generation as knowledge-wise text generation process. Our framework integrates advanced Retrieval-Augmented Generation (RAG) pipelines to ensure the generation of trustworthy counter-speech for 8 main target groups identified in the hate speech literature, including women, people of colour, persons with disabilities, migrants, Muslims, Jews, LGBT persons, and other. We built a knowledge base over the United Nations Digital Library, EUR-Lex and the EU Agency for Fundamental Rights, comprising a total of 32,792 texts. We use the MultiTarget-CONAN dataset to empirically assess the quality of the generated counter-speech, both through standard metrics (i.e., JudgeLM) and a human evaluation. Results show that our framework outperforms standard LLM baselines and competitive approach, on both assessments. The resulting framework and the knowledge base pave the way for studying trustworthy and sound counter-speech generation, in hate speech and beyond.</div
A Categorical Model of Computation for Stateful Stream Programming
No full textWe define a categorical model of computation that mixes contextual and effectful computations, and give a compositional and executable semantics for stateful stream programs. We leverage the Yang-Baxter equation to define a new distributive law between certain monad and comonad. We use the Kleisli category of the resulting distributive law to define a semantics for a minimal yet powerful calculus of imperative stream processing functions. Along the way, we provide illustrative examples to clarify key concepts of the calculus and demonstrate the applicability of its formalization in a stateful implementation of the infinite sieve of Eratosthenes using the Rocq proof assistant.</div
Bug dans plusieurs solveurs de programmation linéaire en nombres entiers
No full textNational audienceLa programmation linéaire en nombre entiers (PLNE) est de plus en plus utilisée au-delà de la communauté de Recherche Opérationnelle. Des chercheurs et chercheuses en conception matérielle ou en biologie se sont notamment emparé·e·s de ces outils pour leurs applications. Les solveurs de PLNE sont ainsi souvent utilisés comme des boîtes noires, sans connaissances particulières quant à leur fonctionnement interne ou de leurs nombreux paramètres.Avec cette présentation nous discutons des bugs potentiels que les utilisateurs et utilisatrices de solveurs peuvent rencontrer. Nous montrons avec l'exemple d'un modèle produit par CarveMe que certains bugs se produisent et peuvent passer inaperçus. Nous tentons ensuite d'en comprendre l'origine sans entrer dans le solveur et d'en éviter la survenue depuis l'étape de modélisation et paramétrage du solveur