HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Tailoring the composition and deposition kinetics of RF-sputtered BaTiO3 thin films through working pressure and target-substrate distance adjustments
No full textInternational audienc
New directions for interconnector research: drawing from social sciences and humanities perspectives to explore the Celtic Interconnector
No full textInternational audienceThe current discourse on interconnectors primarily centers on the technical and economic aspects necessary for delivering a stable grid infrastructure powered by renewable sources and for integrating energy markets. This article, therefore, explores opportunities to broaden definitions of energy grid interconnectivity beyond the techno-economic sphere. It considers multidisciplinary perspectives and presents novel exploratory viewpoints from the social sciences and humanities. It examines ideas of interconnection by drawing on the Celtic Interconnector, an Irish-French initiative, to explore the cultural, historical, political, and geographical dimensions of interconnectivity. Insights are derived from two workshops with academics in Ireland and France, encouraging a more contextual understanding of energy interconnections beyond their physical and economic dimensions. The article builds on these insights to set out an agenda for future research and reflect on frames of reference for describing, analysing, and engaging with emerging interconnector processes and the multiple stakeholders involved
CausalProfiler: Generating Synthetic Benchmarks for Rigorous and Transparent Evaluation of Causal Machine Learning
No full textInternational audienceCausal machine learning (Causal ML) aims to answer "what if" questions using machine learning algorithms, making it a promising tool for high-stakes decision-making. Yet, empirical evaluation practices in Causal ML remain limited. Existing benchmarks often rely on a handful of hand-crafted or semi-synthetic datasets, leading to brittle, non-generalizable conclusions. To bridge this gap, we introduce CausalProfiler, a synthetic benchmark generator for Causal ML methods. Based on a set of explicit design choices about the class of causal models, queries, and data considered, the CausalProfiler randomly samples causal models, data, queries, and ground truths constituting the synthetic causal benchmarks. In this way, Causal ML methods can be rigorously and transparently evaluated under a variety of conditions. This work offers the first random generator of synthetic causal benchmarks with coverage guarantees and transparent assumptions operating on the three levels of causal reasoning: observation, intervention, and counterfactual. We demonstrate its utility by evaluating several state-of-the-art methods under diverse conditions and assumptions, both in and out of the identification regime, illustrating the types of analyses and insights the CausalProfiler enables
Feedback-MPPI: Fast Sampling-Based MPC via Rollout Differentiation – Adios low-level controllers
No full textInternational audienceModel Predictive Path Integral control is a powerful sampling-based approach suitable for complex robotic tasks due to its flexibility in handling nonlinear dynamics and non-convex costs. However, its applicability in real-time, highfrequency robotic control scenarios is limited by computational demands. This paper introduces Feedback-MPPI (F-MPPI), a novel framework that augments standard MPPI by computing local linear feedback gains derived from sensitivity analysis inspired by Riccati-based feedback used in gradient-based MPC. These gains allow for rapid closed-loop corrections around the current state without requiring full re-optimization at each timestep. We demonstrate the effectiveness of F-MPPI through simulations and real-world experiments on two robotic platforms: a quadrupedal robot performing dynamic locomotion on uneven terrain and a quadrotor executing aggressive maneuvers with onboard computation. Results illustrate that incorporating local feedback significantly improves control performance and stability, enabling robust, high-frequency operation suitable for complex robotic systems.</div
Data Stream Processing Effectiveness in Heterogeneous Computing Environments
No full textInternational audienceDistributed Stream Processing (DSP) systems face significant challenges when deployed in heterogeneous environments such as fog computing infrastructures. While these systems were designed for homogeneous cloud environments, their deployment across devices with varying capabilities introduces performance considerations that are not well understood. We present a systematic evaluation of Apache Flink's performance when scaling stateful operations across heterogeneous hardware. Through extensive experiments on a geodistributed testbed comprising bare-metal servers, virtual machines, and Raspberry Pi devices, we examine how different TaskManager configurations affect throughput and resource utilization. Our results reveal that Flink's default configuration mechanisms can lead to significant performance limitations in heterogeneous environments, particularly due to its resource-agnostic task assignment and hash-based partitioning strategy. We identify optimal resource configurations for different node types and quantify the performance impact of various deployment strategies. These findings highlight important considerations for DSP system design in increasingly heterogeneous computing environments
A review of corporate climate ratings: Assessing divergence from scientific expectations
No full textInternational audiencePrivate investment and consumption choices serve as a major driver to push companies to cut their greenhouse gas emissions and align their activities with the goal of reaching global Net Zero. Sustainability scores, labels and rankings have helped guide these decisions since the 1990s. They thus have significant leverage on defining future energy consumption and production, and, more generally, the upcoming low-carbon economy. Yet, such tools are now coming under increasing scholarly criticism. In this context, this study offers a review of issues raised by scholars, an inventory of climate-related scores, labels and ranking providers and their offerings, and an assessment of scores against best-in-class practices for each issue. The concerns raised in the scientific literature are related to the accuracy, reliability, and fairness of the tools, and whether they are effective in driving corporate action. Tool providers were found to use a diversity of business models, methodologies, and definitions of corporate climate performance. Despite some variability across tools and concerns, tools remain generally opaque and poorly aligned with scientific expectations. While corporate climate performance systems typically address indirect impacts and industry and size specificities; they rarely use standardized, verified inputs, and transparent, science-based weightings. Investors, corporations, and researchers can use our results to inform their choice of information providers, and regulators might take interest in the snapshot we provide on the maturity of the corporate climate performance measurement market. This paper aims to initiate improvements in the design of sustainability information systems
Assessing the Limitations of Activation Clipping for Fault Mitigation in Vision and Language Transformers
No full textInternational audienceDeep Neural Network (DNN) activation clipping is a well-established method for mitigating hardware-induced faults during inference. Clipping constrains activations to predefined ranges and filtering corrupted values, including NaN and Inf. However, high variability in activation distributions, both across and within layers of a model, makes it challenging to define fixed ranges that reliably capture corrupted values beyond extreme cases. In this study, we show that activation clipping alone is insufficient to protect vision and language Transformers from inference faults, and that inputs exhibit markedly different levels of fault sensitivity depending on the model's confidence. Our experiments show that the missclassification rate can reach 10.28% even with per-layer clipping is employed