HAL-ENS-LYON
Not a member yet
104280 research outputs found
Sort by
Equivariant Splitting: Self-supervised learning from incomplete data
No full textSelf-supervised learning for inverse problems allows to train a reconstruction network from noise and/or incomplete data alone. These methods have the potential of enabling learning-based solutions when obtaining ground-truth references for training is expensive or even impossible. In this paper, we propose a new self-supervised learning strategy devised for the challenging setting where measurements are observed via a single incomplete observation model. We introduce a new definition of equivariance in the context of reconstruction networks, and show that the combination of self-supervised splitting losses and equivariant reconstruction networks results in unbiased estimates of the supervised loss. Through a series of experiments on image inpainting, accelerated magnetic resonance imaging, sparse-view computed tomography, and compressive sensing, we demonstrate that the proposed loss achieves state-of-the-art performance in settings with highly rank-deficient forward models. The code is available at https://github.com/vsechaud/Equivariant-Splittin
Opportunities from energy-loss near-edge fine structure analysis to track chemical and structural damage in zircon
No full textInternational audienceZircon (ZrSiO4) is the oldest-known mineral of Earth and an ubiquitous silicate in geochronology. More specifically, the accumulation of alpha decay damage in zircon over time significantly affects its physical and chemical properties, and can lead to a disturbance of the ages measured in this mineral. Therefore, analytical tools that enable comprehensive structural and chemical information at the nanoscale in this compound are highly sought after. In this context, we explore the electron energy-loss fine structures resulting from the excitation of O1s and Si2p core electrons in zircon, which are interpreted from ab initio calculations in a single-particle framework. An excellent agreement is obtained between the experimental and calculated fine structures, emphasizing the large distortion of the final electronic states induced by the core-hole potential. The O-K edge is particularly rich in information, with intense peaks dominated by O2p - Zr4d and O2p - Si3sp hybrids. This work suggests that the near-edge structures from the O1s and Si2p excitations accessible from electron energy-loss spectroscopy or X-ray absorption spectroscopy could be used as tools to follow, interpret and understand structural and chemical modifications in zircon subject to natural radiation damage. We illustrate the potential of this approach through the evolution of near-edge fine structures in a zircon sample that exhibits a locally amorphized zone formed by ultrafast laser excitation
Untangling GPU Power Consumption: Job-Level Inference in Cloud Shared Settings
No full textInternational audienceAs the demand for AI-driven workloads increases, the energy consumption of Graphics Processing Units (GPUs) devices has come under intense scrutiny, particularly in hyperscale data centers where large numbers of accelerators are centralized and leased to diverse clients.In the context of cloud hyperscalers, GPUs power monitoring presents several challenges that vary depending on the product offered. The monitoring capabilities of physical devices may be limited or even absent for some products. However, given the substantial energy demands of GPUs, power monitoring is essential for both cloud providers and clients. Operators require tools to manage power distribution effectively, such as balancing workloads across Power Distribution Units (PDUs), while clients need visibility into power usage to optimize their workloads for energy efficiency.To address these challenges, we propose methods for estimating the energy consumption of jobs running on GPU devices in cloud environments, spanning from shared and managed offerings like ML-as-a-Service (MLaaS) to less managed products (e.g., Infrastructure-as-a-Service (IaaS)). Our models demonstrate the benefits of sharing GPUs for small AI workloads, as well as the current sub-optimal utilization of GPUs in cloud hyperscalers, based on insights from an IaaS GPU cluster
Les enjeux communicationnels et interactionnels de l’analyse pragma-énonciative des points de vue (2) : pour une approche holistique des points de vue dans des textes complexes et complets
No full textInternational audienc
Intermediation Platforms and Geopolitical Asymmetries, Lessons from a Pandemic
No full textInternational audienceTo a large extent, and particularly before vaccines became available, human societies owed their resilience during the COVID-19 pandemic to non-pharmaceutical interventions, i.e. social distancing combined with more invasive digital systems. In this paper, we consider the digital applications developed during 2020-2021, the two first years of the pandemic. We introduce a typology based on the services offered and the data flows they require, between both public and private actors. A detailed timeline of these developments shows that countries’ strategies have evolved in strong coherence with their overall digital policy. Our study demonstrates that this exogenous crisis has reinforced the critical role of intermediation platforms for maintaining society’s essential functions. Their increased centrality has contributed to intensifying information asymmetries and power imbalances, already at stake before the pandemic, both between platforms and states, as well as between countries, leading to new geopolitical equilibria
Spin-filament alignments to unravel galaxy evolution and model intrinsic alignments
No full textInternational audienceBy the 2040s, several all-sky surveys will have transformed our view of the large-scale structure. However, one of the major outstanding questions in astrophysics will remain: understanding how galaxies acquire and evolve their angular momentum and how this connects to the cosmic web. Measuring the alignments between galaxy spins and cosmic filaments across cosmic time, and understanding what this reveals about galaxy evolution, requires surveys that also characterise intrinsic alignments, i.e. correlations in galaxy shapes produced by the cosmic web itself rather than by lensing. Intrinsic alignments are a major source of systematic error in weak-lensing measurements of the fundamental parameters of the Universe. Addressing both questions together will necessitate new types of MOS surveys that combine kinematic information with high-completeness redshifts down to at least 24-25mag. To achieve our science goals, we require a new generation of wide-field spectroscopic facilities that can obtain spin-filament alignment measurements for millions of galaxies while simultaneously delivering sub-Mpc resolution of the cosmic web and spatially-resolved kinematics required to map the spin-filament connection at the level of individual galaxies within their local cosmic environment. Such a program would provide a unique legacy survey of galaxies and cosmic structures from kiloparsec to megaparsec scales, establishing ESO's leadership in bridging the physics of galaxy evolution with the systematic-control requirements for Stage-IV cosmological surveys
Biology and Clinical Management of Non-V600 BRAF Alterations in NSCLC
No full textInternational audienc
Le « eux » et le « nous » : rapports conflictuels inter-groupes chez les animaux sociaux
No full textInternational audienceOne of the great constants in the history of human societies is the opposition between “us,” which embodies all conceivable positive values, and “them,” which is associated with everything perceived as negative. Contempt or rejection of “the other” (clan, tribe, race, nation, region, class, caste, ethnic or religious group, family) is the principle behind all ethnocentrism. But the opposition between "us" and "them" is only the extension, in the cultural-symbolic order specific to our species, of a general defense mechanism present throughout the animal kingdom, from eusocial insects to primates: defense of the “close” or the “same” (whether related or unrelated) against what is perceived as distant, different, foreign, or outside one's own group. In the case of human societies, “us” and “them” can take very different forms, as groups are built on cultural foundations. While they have succeeded in bringing together millions (even billions) of individuals into coherent and relatively peaceful social units, human societies have constantly multiplied the types of us/them opposition, and therefore the potential conflicts between groups.L'un des grands invariants dans l'histoire des sociétés humaines réside dans l'opposition entre un « nous », chargé de toutes les valeurs positives imaginables, et un « eux », associé à tout ce qui est perçu comme négatif. Le mépris ou le rejet de l'« autre » (clan, tribu, ethnie, race, nation, région, groupe religieux, etc.) est le principe de tout ethnocentrisme. Mais l'opposition eux/nous n'est que le prolongement, dans l'ordre culturel-symbolique propre à notre espèce, d'un mécanisme général de défense présent dans l'ensemble du règne animal, des insectes eusociaux aux primates : défense du « proche » ou du « même » (apparenté ou nonapparenté) par rapport à ce qui est perçu comme lointain, différent, étranger, extérieur à son propre groupe. Dans le cas des sociétés humaines, le « nous » et les « eux » peuvent prendre des formes très variables dans la mesure où les groupes se construisent sur des bases culturelles. En même temps qu'elles ont réussi à rassembler des millions (voire des milliards) d'individus dans des unités sociales cohérentes et relativement pacifiées, les sociétés humaines n'ont cessé de multiplier les types d'opposition eux/nous, et donc les conflits possibles entre les groupes