INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Clarify status of non-returning functions with respect to function attributes
No full textThe current wording in 6.7.13.8.1 has lead to misunderstandings about the status of function calls with respect to the [[reproducible]] and [[unsequenced]] attributes for the case that such an attributed function does not return. It seems that the corresponding gcc attributes __attribute__((pure)) and __attribute__((const)) assumed that such calls always return, without clearly documenting that expectation, nor by documenting its reach
Another daemon: waiting for condition variables
No full textAccepted for integration into the next standard for the C programming languageThe two wait funtions for conditional variables use non-binding terminology to describe a requirement. For example they claim that these functions would “require” the passed-in mutex to be locked. The use of such wording is not binding normatively
Autonomic Resource Harvesting in HPC: Control Methods and their Reusability
No full textInternational audienceHigh Performance Computing (HPC) systems are subject to dynamical variations occurring in e.g., jobs execution duration, I/O quantity, network consumption. Adapting to these unpredictable variations requires using autonomic management in an online feedback loop. The introduction of control theory methods allows for the design of well-founded autonomic managers. Choosing the relevant approach is daunting due to the variety of existing controllers. The criteria are of different natures, involving performance and efficiency, but also required expertise in control theory, and reusability or portability between sub-systems. Therefore, there is a need for comparative studies to assist designers choices.We consider the problem of resource harvesting in HPC systems, where scheduling often leaves resources idle. Our approach controls -through a feedback loop -the injection of small jobs in order to maximize the resources' usage. The control problem is to manage the trade-off between harvesting and performance, in a reusable manner. We study how reusability relates to the adaptivity and robustness properties in control. We illustrate our approach with the classic Proportional-Integral-Derivative (PID) control, its upgrade as adaptive control, and Model-Free Control (MFC). We target CiGri, a system harvesting idle resources in a computing grid. We perform experimental evaluation and compare performance and reusability. Trade-offs are found on different criteria: while adaptive control is largely portable, its design complexity is significant for non-experts; PID control has good nominal performance, yet its portability is limited; MFC requires few competences to be used, but cannot provide strong guarantees
The Harder Path: Last Iterate Convergence for Uncoupled Learning in Zero-Sum Games with Bandit Feedback
No full textInternational audienceWe study the problem of learning in zero-sum matrix games with repeated play and bandit feedback. Specifically, we focus on developing uncoupled algorithms that guarantee, without communication between players, the convergence of the last-iterate to a Nash equilibrium. Although the non-bandit case has been studied extensively, this setting has only been explored recently, with a bound of (T-1/8) on the exploitability gap. We show that, for uncoupled algorithms, guaranteeing convergence of the policy profiles to a Nash equilibrium is detrimental to the performance, with the best attainable rate being Ω(T -1/4 ) in contrast to the usual Ω(T-1/8) rate for convergence of the average iterates. We then propose two algorithms that achieve this optimal rate up to constant and logarithmic factors. The first algorithm leverages a straightforward trade-off between exploration and exploitation, while the second employs a regularization technique based on a two-step mirror descent approach
An efficient neural network-based surrogate model for predicting static gear contact conditions
No full textInternational audienceGears are an essential component of numerous mechanical systems across a wide range of engineeringapplications. However, they may be associated to high levels of radiated noise which can limit their use.Accurately predicting this noise is of paramount importance for the design, optimization and health mon-itoring of gear transmissions. System identification is therefore needed to reach a sufficiently high levelof accuracy. However, this usually comes at the cost of high computational burden. Using traditionalmodeling assumptions, it is widely accepted that the radiated noise stems from the dynamic response ofthe gears which is itself induced by the static transmission error (STE) and time-varying mesh stiffness.These physical quantities are governed by the local contact conditions between the gear teeth. An accu-rate computation of these physical quantities is therefore crucial. However, this is a difficult problem asgear contact resolution is intrinsically nonlinear and multiscale. Even considering simplifying assump-tions, the computation of this physical quantities entails a significant computational effort when coupledto optimization procedures. In this work, we introduce an efficient neural network-based surrogate modelfor predicting static gear contact conditions in near real time in order to facilitate the identification andoptimization of mechanical systems equipped with geared systems
Sample-efficient decoding of visual stimuli from fMRI through inter-individual functional alignment
No full textInternational audienceDeep learning is leading to major advances in the realm of brain decoding from functional Magnetic Resonance Imaging (fMRI). However, the large inter-individual variability in brain characteristics has constrained most studies to train models on one participant at a time. This limitation hampers the training of deep learning models, which typically requires very large datasets. Here, we propose to boost brain decoding of videos and static images across participants by aligning brain responses of training and left-out participants. Evaluated on a retrieval task, compared to the anatomically-aligned baseline, our method halves the median rank in out-of-subject setups in low-data regimes. It also outperforms classical within-subject approaches when fewer than 100 minutes of data is available for the tested participant. Furthermore, we show that our alignment framework handles multiple subjects, which improves accuracy upon classical single-subject approaches. Finally, we show that this method aligns neural representations in accordance with brain anatomy. Overall, this study lays the foundations for leveraging extensive neuroimaging datasets and enhancing the decoding of individual brains when a limited amount of brain-imaging data is available.</div
MaTOS: Machines à Traduire pour Ouvrir la Science
No full textInternational audienceThis paper is a short presentation of MaTOS (Machine Translation for Open Science), a project focusing on the automatic translation of scholarly documents. Its main aims are (a) to develop resources (term lists and corpora) for high-quality machine translation, (b) to study methods for translating complete, structured documents in a cohesive and consistent manner, (c) to propose novel metrics to evaluate machine translation in technical domains. Publications and resources are available on the project web site: https://anr-matos.fr.Cet article présente brièvement MaTOS (Machines à Traduire pour Ouvrir la Science), un projet axé sur la traduction automatique de documents scientifiques. Ses principaux objectifs sont (a) de développer des ressources (listes de termes et corpus) pour une traduction automatique de haute qualité, (b) d'étudier des méthodes permettant de traduire des documents complets et structurés de manière cohérente et homogène, (c) de proposer de nouveaux indicateurs pour évaluer la traduction automatique dans les domaines techniques. Les publications et les ressources sont disponibles sur le site web du projet : https://anr-matos.fr
Accurate frictional contact algorithms for the numerical exploration of the mechanics of fibrous assemblies
No full textInternational audienceThe goal of this work is to explore novel algorithms to enhance the predictability of fibre assembly simulators, without sacrificing the complexity of the target scenarios. Our work relies on the super-helix curvature-based discrete model for Kirchhoff elastic rods (1), coupled with the non-smooth so-bogus (2) solver for frictional contact resolution. While previously validated at the geometrical level in both 2D and 3D configurations (3), the contact forces produced by the simulations can exhibit spurious jumps, inconsistent with smoothly sliding configurations. Starting from a detailed analysis of a simulation of the three point bending test, we show that these jumps are not specific to our model, but result from the use of low order contact detection methods. We propose new algorithms to achieve efficient and accurate high-order contact detection (4), thereby retrieving artefact-free forces, as expected from our high-order discretisation
Effective Asymptotics of Combinatorial Systems
No full textAnalytic combinatorics studies asymptotic properties of families of combinatorial objects using complex analysis on their generating functions. In their reference book on the subject, Flajolet and Sedgewick describe a general approach that allows one to derive precise asymptotic expansions starting from systems of combinatorial equations. In the situation where the combinatorial system involves only cartesian products and disjoint unions, the generating functions satisfy polynomial systems with positivity constraints for which many results and algorithms are known. We extend these results to the general situation. This produces an almost complete algorithmic chain going from combinatorial systems to asymptotic expansions. Thus, it is possible to compute asymptotic expansions of all generating functions produced by the symbolic method of Flajolet and Sedgewick when they have algebraic-logarithmic singularities (which can be decided), under the assumption that Schanuel's conjecture from number theory holds. That conjecture is not needed for systems that do not involve the constructions of sets and cycles
Leveraging Expert Usage to Speed up LLM Inference with Expert Parallelism
No full textInternational audienceLarge language models have become indispensable for many text-processing applications. Their inference, i.e. their use to generate text, is a time-consuming task since tokens have to be generated one after the other, even if the computational load has been reduced by model sparsification, e.g. by using a Mixture of Experts (MoE) models. In the MoE context, a subset of experts is selected at each stage. Note that not all subsets of experts (pairs of experts in most cases) in a given layer have the same probability of being selected. When experts are mapped to different GPUs, there is a risk of load imbalance if the selected experts end up on a small number of GPUs. This paper proposes to leverage this heterogeneity in expert usage to map experts of popular subsets onto distinct GPUs, allowing them to be processed in parallel and thus reducing the time needed for inference. Even though this mapping problem is NP-complete, it is possible to design simple greedy strategies that significantly reduce the need for sequential expert processing. Our proof-ofconcept confirms that our mapping strategies effectively reduce inference time on the Mixtral model