INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Choice Trees: Representing and Reasoning About Nondeterministic, Recursive, and Impure Programs in Rocq
No full textInternational audienceThis paper introduces Choice Trees (CTrees), a monad for modeling nondeterministic, recursive, and impure programs in Rocq. Inspired by Xia et al.'s ITrees, this novel data structure embeds computations into coinductive trees with three kinds of nodes: external events, internal steps, and delayed branching. This structure allows us to provide shallow embedding of denotational models with nondeterministic choice in the style of ccs, while recovering an inductive LTS view of the computation. CTrees leverage a vast collection of bisimulation and refinement tools well-studied on LTSs, with respect to which we establish a rich equational theory.We connect CTrees to the ITrees infrastructure by showing how a monad morphism embedding the former into the latter permits using CTrees to implement nondeterministic effects. We demonstrate the utility of CTrees by using them to model concurrency semantics in two case studies: ccs and cooperative multithreading
A Theoretical Framework for Grokking: Interpolation followed by Riemannian Norm Minimisation
No full textInternational audienceWe study the dynamics of gradient flow with small weight decay on general training losses F : R d → R. Under mild regularity assumptions and assuming convergence of the unregularised gradient flow, we show that the trajectory with weight decay λ exhibits a two-phase behaviour as λ → 0. During the initial fast phase, the trajectory follows the unregularised gradient flow and converges to a manifold of critical points of F. Then, at time of order 1/λ, the trajectory enters a slow drift phase and follows a Riemannian gradient flow minimising the ℓ2-norm of the parameters. This purely optimisation-based phenomenon offers a natural explanation for the grokking effect observed in deep learning, where the training loss rapidly reaches zero while the test loss plateaus for an extended period before suddenly improving. We argue that this generalisation jump can be attributed tothe slow norm reduction induced by weight decay, as explained by our analysis. We validate this mechanism empirically on several synthetic regression tasks
Positivity-preserving well-balanced pampa schemes with global flux quadrature for one-dimensional shallow water models
No full textWe present a novel hydrostatic and non-hydrostatic equilibria preserving Point-Average-Moment PolynomiAl-interpreted (PAMPA) method for solving the one-dimensional hyperbolic balance laws, with applications to the shallow water models including the Saint-Venant system with the Manning friction term and rotating shallow water equations. The idea is based on a global flux quadrature formulation, in which the discretization of the source terms is obtained from the derivative of and additional flux function computed via high order quadrature of the source term. The reformulated system is quasi-conservative with global integral terms computed using Gauss-Lobatto quadrature nodes. The resulting method is capable of preserving a large family of smooth moving equilibria: supercritical and subcritical flows, in a super-convergent manner. We also show that, by an appropriate quadrature strategy for the source, we can exactly preserve the still water states. Moreover, to guarantee the positivity of water depth and eliminate the spurious oscillations near shocks, we blend the high-order PAMPA schemes with the first order local Lax-Friedrichs schemes using the method developed in [2]. The first-order schemes are designed to preserve the still water equilibria and positivity of water height, as well as to deal with wet-dry fronts. Extensive numerical experiments are tested to validate the advantages and robustness of the proposed scheme
Estimation de la consommation d'énergie des applications cloud-native
No full textThis thesis introduces an empirical framework for accurately estimating the energy footprint of cloud-native applications through two complementary studies. First, we demonstrate that CPU core's static power can account for up to 12% of package consumption and propose a three-step approach (data generation, offline thermal correction, static power estimation) that ensures transparency and reproducibility. Second, our custom memory-intensive stress test and correlation analysis pinpoint Core/Uncore frequencies, data volume, and parallelism degree as the main drivers of cache energy and power, enabling nonlinear regression models with R2 ≥95%. These insights lay the groundwork for an enhanced RAPL-like interface for cache monitoring and dynamic energy management strategies in cloud-native environments.Ce travail propose un cadre expérimental pour estimer finement la consommation énergétique des applications cloud-native. D’une part, nous démontrons que la puissance statique des cœurs CPU peut représenter jusqu’à 12% de la consommation totale du package et décrivons une méthodologie en trois étapes (génération de données, correction thermique hors ligne, estimation de la puissance statique) garantissant la reproductibilité et la transparence. D’autre part, notre logiciel de tests mémoire personnalisé et notre analyse de corrélation identifient la fréquence Core/Uncore, le volume de données et le degré de parallélisme comme principaux leviers de la consommation énergétique des caches, avec des modèles de régression non linéaires atteignant : R2 ≥ 95%. Ces résultats ouvrent la voie à l’extension de l’interface RAPL pour la supervision des caches et à des stratégies dynamiques d’optimisation énergétique dans un environnement cloud-native
LOGLO-FNO: Efficient Learning of Local and Global Features in Fourier Neural Operators
No full textInternational audienceModeling high-frequency information is a critical challenge in scientific machine learning. For instance, fully turbulent flow simulations of the Navier-Stokes equations at Reynolds numbers 3500 and above can generate high-frequency signals due to swirling fluid motions caused by eddies and vortices. Faithfully modeling such signals using neural networks depends on the accurate reconstruction of moderate to high frequencies. However, it has been well known that deep neural nets exhibit the so-called spectral or frequency bias towards learning low-frequency components. Meanwhile, Fourier Neural Operators (FNOs) have emerged as a popular class of data-driven models in recent years for solving Partial Differential Equations (PDEs) and for surrogate modeling in general. Although impressive results have been achieved on several PDE benchmark problems, FNOs often perform poorly in learning non-dominant frequencies characterized by local features. This limitation stems from the spectral bias inherent in neural networks and the explicit exclusion of high-frequency modes in FNOs and their variants. Therefore, to mitigate these issues and improve FNO's spectral learning capabilities to represent a broad range of frequency components, we propose two key architectural enhancements: (i) a parallel branch performing local spectral convolutions and (ii) a high-frequency propagation module. Moreover, we propose a novel frequency-sensitive loss term based on radially binned spectral errors. This introduction of a parallel branch for local convolutions reduces the number of trainable parameters by up to 50% while achieving the accuracy of the baseline FNO that relies solely on global convolutions. Moreover, our findings demonstrate that the proposed model improves the stability over longer rollouts. Experiments on six challenging PDE problems in fluid mechanics, wave propagation, and biological pattern formation, and the qualitative and spectral analysis of predictions, show the effectiveness of our method over the state-of-the-art neural operator families of baselines
Incorporating 3D-Rendered Materials in Visualization
No full textInternational audienceWe investigate how 3D-rendered materials can support expressive forms of information visualization. We introduce an early snapshot of our design space, describing how inherent material properties and their state or structural transformations can be used as visual channels or simply as contextual attributes for sensory activation. We explore the potential of rendered materials to evoke emotional engagement, curiosity, aesthetic pleasure, and crossmodal sensory experiences
Botascopia: a digital setup for generating low tech plant field guides
No full textPlant field guides are portable reference books or digital tools designed to help various users identify plant species in a given area and find out about their morphology, ecology and usage. We present Botascopia, a combination of digital tools that automatically produce low-tech field guides in the form of printed paper booklets containing species description sheets associated with an identification key. These are analogous to floras adapted to local ecosystems, such as gardens, parks, school yards or campuses. Botascopia contains a participatory knowledge base with detailed descriptions of plant species. It uses computational languages to express different perspectives on plants held by expert botanists and novice observers, for example, and the translations between them. This setup therefore offers a solution to the usual challenge of balancing precise vocabulary with making botanical knowledge accessible beyond expert communities. We tested this approach during botanical field sessions with diverse audiences, providing them with personalised, automatically generated printed booklets. This experiment explores how combining low- and high-tech approaches can contribute to the study of the relationships between technology, humans, and plants in the Anthropocene
Functional central limit theorem for topological functionals of Gaussian critical points
No full textWe consider Betti numbers of the excursion of a smooth Euclidean Gaussian field restricted to a rectangular window, in the asymptotics where the window grows to R^d . With motivations coming from Topological Data Analysis, we derive a functional Central Limit Theorem where the varying argument is the thresholding parameter, under assumptions of regularity and covariance decay for the field and its derivatives. We also show fixed-level CLTs coming from martingale based techniques inspired from the theory of geometric stabilisation, and limiting non-degenerate variance
Childhood trauma affects speech and language measures in patients with major depressive disorder during clinical interviews
No full textInternational audienceBackground: Speech analysis has shown significant promise as a potential biomarker for depression. However, no studies to date have examined the impact of childhood trauma on speech and language patterns in individuals with depression. This study aims to explore the relationship between vocal characteristics and depressive symptoms, while also assessing how childhood trauma may shape these patterns. Methods: 27 participants with a major depressive episode were included. The severity of depression was assessed using the Montgomery & Asberg Depression Rating Scale (MADRS) and the Beck Depression Inventory II. Childhood trauma was measured using the Childhood Trauma Questionnaire. Speech recordings from the MADRS semi-structured interview and a free clinical interview were analyzed using speaker diarization, automatic speech recognition, and feature extraction. Results: Several acoustics features were significantly associated with depression severity. Correlation analysis revealed that greater depression severity was linked to shorter, less diverse speech, characterized by fewer words, fewer semantic clusters, and reduced articulatory effort. In contrast, childhood trauma was positively associated with distinct speech characteristics. Higher trauma load was associated with richer, longer, and more syntactically complex speech. Additionally, utterances were shorter, with more frequent shifts between semantic clusters, reflecting a more fragmented speech pattern influenced by traumatic load. Conclusion: Our study highlights the influence of childhood trauma on vocal and linguistic characteristics of patients with depression. Automated language analysis offers the possibility to identify biomarkers of traumatic load in patients with depression. This could improve diagnostic accuracy, guide therapeutic management and monitor clinical progress
A view of Research and Standardisation: the oneM2M IoT standard by ETSI
No full textInternational audienceMy personal view on how research and standardisation in IoT should go hands in hand