HAL-MINES ParisTech
Not a member yet
27348 research outputs found
Sort by
EAS-Sim: A Framework and its Methodology for the Co-Design of Multi-Objective, Energy-Aware Schedulers for AI Clusters
No full textInternational audienceThe explosive growth of large-scale DL models has made energy consumption a first-order operational cost and constraint in modern HPC datacenters. Existing DL schedulers, however, are largely single-objective and energy oblivious, struggling to balance the competing demands of performance, fairness, and QoS. To address this flaw, we propose a methodology for the co-design of multi-objective and energy-aware schedulers together with the associated simulation framework, the so-called EAS-Sim. Our methodology stands as a systematic approach to enhance SOTA scheduling heuristics with energy-efficiency objectives.Using our framework, we design and evaluate four novel and malleable job schedulers. Our flagship energy-aware policy, called Zeus, establishes a new Pareto-optimal frontier and reduces total energy consumption by up to 8-10% compared to the SOTA performance scheduler Pollux with no statistically significant loss in system throughput. We extend the methodology with fairness (resource equity), thus producing Hades, which achieves the same level of fairness as Themis with a 7-8% energy reduction. We further prove that robust SLA can be guaranteed only through preemption. Another policy we propose, called AuraChronos, achieves a nearly perfect deadline miss rate of less than 1%, thus more than x20 improvement over non-preemptive policies. Finally, we validate Charon, another policy from us that can operate successfully under a tight power budget. EAS-Sim is available as open-source on GitHub (https://github.com/HPC-CRI/EAS-Sim)
Bubble growth model and validation of chemical foaming modelling of rubber
No full textInternational audienceIn this work, we present all the essential concepts and parameters required to model thechemical foaming process of an EPDM rubber. The main approach proposed here is thedevelopment and application of a bubble growth model.In his thesis, Juan Itriago has proposed a kinetic model describing the density evolutionbased on the two kinetic reactions: namely the gas generation due to the OBSHdecomposition and the vulcanisation reaction that freezes the cellularised structure. Thismodel is built from experimental dilatometry tests: free foaming measurements in an ovenand dilatometry tests under pressure in a tensile testing machine. Unfortunately, therheological behaviour of rubber and the diffusivity of gas are not taken into account in thismodel.Therefore, the validity of this simple kinetic model is checked by means of a classical bubblegrowth model (BGM) based on Amon and Denson’s work but modified to take into accountthe particularities of working with a chemically blown elastomer. In particular, the use of achemo-rheological model to describe the effect of the cross-linking reactions associated withthe vulcanization process on the viscosity of the rubber is introduced.This approach considers a single cell formed by a bubble surrounded by a spherical fluidshell containing gas generated by a chemical reaction. As the dissolved gas diffuses towardsthe bubble, the pressure increases, inducing bubble growth. The bubble continues to growuntil the moment when no more gas is available. By applying the conservation of momentum,mass, and energy principles to this process, it is possible to obtain a set of equationsdescribing the evolution of the bubble radius, the pressure, and the concentration gradientinside the polymer. In addition to initial and boundary conditions (such as the initial gasconcentration, bubble density, temperature, and pressure in the surrounding medium),different physical parameters need to be determined or estimated: diffusivity, Henry’sconstant, surface tension of the fluid, and matrix’s viscosity, density and heat capacity.A second part consists in checking the kinetics model by using a macroscale simulation ofthe whole injection molding process, including the polymer flow into the cavity during thefilling step and the evolution of its temperature throughout the molding step. Preliminarycomparisons with some experimental observations will be presented in order to illustrate therole of temperature and the rheology on the injection foaming process</div
Retrieval of Long-Term (1980–2024) Time Series of PM10 Concentration by an Empirical Method: The Paris, Cairo, and New Delhi Case Studies
No full textInternational audiencePluriannual time series of fine particle concentrations suspended in the atmosphere are often lacking. Such data is necessary in evaluating the efficiency of policies aiming to improve air quality in megacities. In this work, a recently developed empirical method is applied over the megacities of Paris, Cairo, and New Delhi. The method utilizes observations of the aerosol optical depth, Angström Exponent, and atmospheric precipitable water as inputs to estimate the PM10. The modeled values validated against their respective reference measurements exhibited the best performance at daily, weekly, and monthly averages when using inputs of the AERONET. When exploiting inputs of the CAMS and MERRA-2 reanalyses, the results were found to be satisfactory with MERRA-2 on the monthly scale. This allows the reconstruction of the variability of the PM10 for the last 45 years. Analysis shows that average annual PM10 concentration has decreased from 40 to 20 µg·m−3 in Paris, increased from 70 to 250 µg·m−3 in New Delhi, and stayed relatively stable (around 100 µg·m−3) in Cairo. Provided that at least one year of PM10 measurements are available to calibrate the empirical method, the method herein is replicable over other megacities around the world
Strongly driven transmon as an incoherent noise source
No full textUnder strong drives, which are becoming necessary for fast high-fidelity operations, transmonscan be structurally unstable. Due to chaotic effects, the computational manifold is no longer wellseparated from the remainder of the spectrum, which correlates with enhanced offset-charge sensitivity and destructive effects in readout. We show here that these detrimental effects can further propagate to other degrees of freedom, for example to neighboring qubits in a multi-qubit system.Specifically, a coherently driven transmon can act as a source of incoherent noise to another circuit element coupled to it. By using a full quantum model and a semiclassical analysis, we perform the noise spectroscopy of the driven transmon coupled to a spectator two-level system (TLS), and we show that, in a certain limit, the interaction with the driven transmon can be modeled as astochastic diffusive process driving the TL
Findings of the First Shared Task for Creole Language Machine Translation at WMT25
No full textInternational audienceEfforts towards better machine translation (MT) for Creole languages have historically been isolated, due to Creole languages' geographic and linguistic diversity. However, most speakers of Creole languages stand to benefit from improved MT for low-resource languages. To galvanize collaboration for Creole MT across the NLP community, we introduce the First Shared Task for Creole Language Machine Translation at WMT25. This Shared Task consists of two systems tracks and one data track, for which we received submissions from five participating teams. Participants experimented with a wide variety of systems and development techniques. Our evaluation campaign gave rise to improvements in MT performance in several languages, and particularly large improvements in new testing genres, though some participants found that reusing subsets of pretraining data for specialized post-training did not yield significant improvements. Our campaign also yielded new test sets for Mauritian Creole and a vast expansion of public training data for two Creole languages of Latin America
Skill‐Driven Data Sampling and Deep Learning Framework for Minute‐Scale Solar Forecasting with Sky Images
No full textInternational audienceAccurate very short‐term solar irradiance forecasting is crucial for optimizing the integration of solar energy into power systems. Herein, an image‐based deep learning framework for minute‐scale solar irradiance prediction is presented. The locally developed model is benchmarked against two commercial forecasting solutions deployed at the same experimental site, demonstrating superior accuracy and adaptability. A key contribution is the introduction of a skill‐driven sampling algorithm based on clear sky index persistence error, which optimizes the training dataset by excluding low‐utility samples while retaining essential physical features like solar zenith and azimuth angles. This algorithm enables the exclusion of up to 30% of the original training data, resulting in ≈16% savings in computational resources without affecting forecast accuracy validated using a test set of 324 991 observations. The model achieves a skill score of 7.63%, significantly outperforming the commercial models, which exhibit negative skill scores under the same conditions
Algorithms for nonsmooth optimization models under distance-to-set penalties
No full textInternational audienc
Hydrogène électrolytique en Europe : progresser en intégrant les incertitudes
No full textInternational audienc
Six Decades of Losses and Gains in Alpha Diversity of European Plant Communities
No full textInternational audienceBiodiversity change forecasts rely on long‐term time series, but such data are often scarce in space and time. Here, we interpolated spatiotemporal changes in species richness using a new method based on machine learning that does not require temporal replication at sites. Using 698,692 one‐time sampled vegetation plots, we estimated trends in vascular plant alpha diversity across Europe and validated our approach against 22,852 independent time series. We found an overall near‐zero net change in species richness between 1960 and 2020. However, species richness generally declined from 1960 to 1980 and increased from 2000 to 2020 across habitats. Declines were most pronounced in forests, but trends varied across habitats and regions, with overall increases at higher latitudes and elevations, and declines or stable trends elsewhere. Our findings demonstrate how data without temporal replication can be used to reveal context‐dependent biodiversity dynamics, underscoring their importance for conservation and management