HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Asymmetrical Latent Representation for Individual Treatment Effect Modeling
No full textConditional Average Treatment Effect (CATE) estimation, at the heart of counterfactual reasoning, is a crucial challenge for causal modeling both theoretically and applicatively, in domains such as healthcare, sociology, or advertising. Borrowing domain adaptation principles, a popular design maps the sample representation to a latent space that balances control and treated populations while enabling the prediction of the potential outcomes. This paper presents a new CATE estimation approach based on the asymmetrical search for two latent spaces called Asymmetrical Latent Representation for Individual Treatment Effect (ALRITE), where the two latent spaces are respectively intended to optimize the counterfactual prediction accuracy on the control and the treated samples. Under moderate assumptions, ALRITE admits an upper bound on the precision of the estimation of heterogeneous effects (PEHE), and the approach is empirically successfully validated compared to the state-of-the-ar
Electronic properties and stability of interstitial oxygen in UO2 grain boundaries: An ab initio study
No full textInternational audienceThe oxidation of UO2 is primarily governed by the diffusion of oxygen through the lattice. Oxygen diffusion is significantly influenced by defects and interfaces, with grain boundaries being particularly relevant in spent nuclear fuel due to their increasing concentration at the periphery of fuel rods. While experimental studies on bulk uranium suggest defects enhance oxygen diffusion, the role of grain boundaries remains contentious, with discrepancies between theoretical predictions of enhanced diffusion and experimental observations. This study employs density functional theory (DFT+U) to investigate the electronic properties and stability of interstitial oxygen in two coincident site lattice grain boundaries, Σ3 {111} and Σ5 {210}, in UO2. We compare stoichiometric and non-stoichiometric grain boundary models, examining their formation energies, defect interactions, and local structural distortions. The interstitial oxygen defects cause an expansion of the oxygen cage and a contraction of the U-O bonds, both in bulk and at the grain boundaries. The Σ3 grain boundary showed potential for defect accumulation, while the Σ5 grain boundary did not demonstrate reduced defect formation energies relative to the bulk. Our findings contribute to the understanding the UO2 oxidation process, in an attempt to address inconsistencies between theoretical and experimental studies on oxygen diffusion in grain boundaries
Evaluating Multi-Channel Input of Sagittal T2-weighted and STIR MRI for Automated Segmentation of Spinal Cord MS Lesions
No full textIdentifying multiple sclerosis lesions in spinal cord MRI is important but challenging, and automated methods have been proposed to reduce inter-rater variability. Although clinicians typically use multiple MR sequences to identify lesions for a given patient, most automated methods use only a single sequence as input. We investigated combining sagittal T2-weighted and STIR sequences using an early fusion U-Net for lesion segmentation, assessing the impact of resampling techniques and U-Net architectures (3D vs. hybrid 2D/3D). Results showed no overall improvement from adding STIR (p=0.17), though this varied among subjects. Furthermore, T2-only models trained on larger datasets outperformed T2+STIR models trained on subsets, indicating that a larger single-sequence dataset is preferable to a smaller multi-sequence one
A lateral porous silicon electrokinetic molecular valve
No full textInternational audienceIn this study, we introduce an Electrokinetic Molecular Valve (EMV) based on Lateral Porous Silicon (LPSi) membranes. The LPSi membranes are fabricated and monolithically integrated into silicon microfluidic chips , featuring an average pore size of 25 nm. Upon proper oxidation, LPSi membranes exhibit a relative perm-selectivity of 48% in physiological solution, comparable to that of Nafion. The LPSi chip is able to extract and concentrate 1.5 fmol of fluorescein from 180 nL into 1.3 nL within 10 minutes, and to achieve a concentration factor of more than 120 at voltages less than 4.2 V. A simplified numerical model is developed to describe the electrokinetic behavior of the EMV. The model exibites good qualitative agreement with experimental results. By varying parameters within this framework (the applied voltage, membrane charge density, background ion concentration, and membrane position), the preconcentration performance of the EMV can be reliably predicted. Distinct from conventional electrokinetic concentrators, the EMV architecture mandates that the entire fluid flow through the LPSi nanochannels. This configuration enables high ion selectivity and low voltage operation, while leveraging the Donnan exclusion effect for precise molecular control, concentration, and release. With continued advancements in electrical insulation and membrane charge density, the proposed EMV holds considerable promise for integration into portable µTAS and biosensors
Perfectly transparent boundary conditions and wave propagation in lattice Boltzmann schemes
No full textSystems of N = 1, 2, . . . first-order hyperbolic conservation laws feature N undamped waves propagating at finite speeds. On their own hand, multi-step Finite Difference and lattice Boltzmann schemes with q = N + 1, N + 2, . . . unknowns involve N "physical" waves, which are aimed at being as closely-looking as possible to the ones of the PDEs, and q-N "numerical-spurious-parasitic" waves, which are subject to their own speed of propagation, and either damped or undamped. The whole picture is even more complicated in the discrete setting-as numerical schemes act as dispersive media, thus propagate different harmonics at different phase (and group) velocities. For compelling practical reasons, simulations must always be conducted on bounded domains, even when the target problem is unbounded in space. The importance of transparent boundary conditions, preventing artificial boundaries from acting as mirrors producing polluting ricochets, naturally follows. This work presents, building on Besse, Coulombel, and Noble [ESAIM: M2AN, 55 (2021)], a systematic way of developing perfectly transparent boundary conditions for lattice Boltzmann schemes tackling linear problems in one and two space dimensions. Our boundary conditions are "perfectly" transparent, at least for 1D problems, as they absorb both physical and spurious waves regardless of their frequency. After presenting, in a simple framework, several approaches to handle the fact that q > N , we elect the so-called "scalar" approach (which despite its name, also works when N > 1) as method of choice for more involved problems. This method solely relies on computing the coefficients of the Laurent series at infinity of the roots of the dispersion relation of the bulk scheme. We insist on asymptotics for these coefficients in the spirit of analytic combinatorics. The reason is two-fold: asymptotics guide truncation of boundary conditions to make them depending on a fixed number of past time-steps, and make it clearduring the process of computing coefficients-whether intermediate quantities can be safely stored using floating-point arithmetic or not. Numerous numerical investigations in 1D and 2D with N = 1 and 2 are carried out, and show the effectiveness of the proposed boundary conditions
Efficient Memory Usage For Edge FaaS Platforms
No full textInternational audienceFunction as a Service (FaaS) is a great fit for data and event processing in Edge environments. These environments are characterized by resource-constrained devices that require efficient memory usage optimizations. Existing optimization so- lutions for memory in high-end clusters such as data centers cannot be used in Edge environments because they either depend on unavailable hardware features such as RDMA or require resource-intensive analysis such as periodic memory scanning, compression, or decompression. In this paper, we introduce Extensible RUNtimes (ERUN), a lightweight mechanism for existing FaaS runtimes aimed at optimizing memory utilization by reducing the memory usage of idle sandboxes (i.e., those awaiting function execution). ERUN operates through two main actions: Shrink and Expand. The Shrink operation unloads libraries and reclaims memory from the sandboxes, while the Expand operation quickly reloads the libraries when a function is executed. The Expand operation leverages an in-memory store that maintains a single instance of discarded libraries on the node. We implement the ERUN mechanism, which can be applied to any runtime environment, in a Python runtime. We extensively evaluated our prototype in a 10-node Edge cluster using 10 popular FaaS functions. The results show that ERUN can reduce idle sandbox memory usage by up to 23.13× and improve the warm-start ratio by 1.38×, while incurring less than 2% overhead on function execution time and energy usage
Less is more: uncompensated gravity torques for intuitive EMG-based assistance with a robotic exoskeleton
No full textDespite extensive investigation on the use of electromyographic (EMG) activity to control active exoskeletons over the past decade, designing intuitive assistive controllers that seamlessly integrate with natural human motor control have yet to be realized. While existing EMG-based controllers often achieve substantial reduction in muscle effort, they frequently incur increased cognitive and attentional load for the user, thereby compromising the overall efficacy of the assistance. Here we introduce a novel EMG-based assistive controller founded upon neuroscience principles, specifically the observation that humans naturally exploit gravity torque to facilitate movement control. Therefore, deviating from conventional compensation strategies, our approach purposely leaves a fraction of the predicted human gravity torque uncompensated so that users can still take advantage of gravity as they would without assistance. Through a load-carrying arm movement task, we show that enabling gravity exploitation improves traditional EMGbased assistance by achieving a significant reduction in muscle effort, while concurrently yielding superior kinematic performance (i.e., faster, smoother movements) and enhanced subjective user experience. These findings demonstrate that integrating principles from neural motor control into assistive controllers allows to implement a favorable tradeoff between muscle effort reduction and functional usability
Promises, Perils, and (Timely) Heuristics for Mining Coding Agent Activity
No full textInternational audienceIn 2025, coding agents have seen a very rapid adoption. Coding agents leverage Large Language Models (LLMs) in ways that are markedly different from LLM-based code completion, making their study critical. Moreover, unlike LLM-based completion, coding agents leave visible traces in software repositories, enabling the use of MSR techniques to study their impact on SE practices. This paper documents the promises, perils, and heuristics that we have gathered from studying coding agent activity on GitHub
Irradiation response during the early stages of alpha radiation damage in mesoporous nanocrystalline ceria films
No full textInternational audienceWe have studied the effect of alpha radiation on mesoporous ceria in the very early stages of damage by electronic energy loss processes. Most of the studies of the radiation effects in nanocrystalline ceria were performed on relatively dense samples and in the ballistic regime of binary collisions. Few studies on ceramic systems have focused on radiation effects in loosely packed nanocrystalline systems with predominantly free surfaces. Free surfaces are believed to be critical for radiation resistance, but few studies specifically investigated radiation-induced changes of structural correlations in such systems. We irradiated mesoporous nanocrystalline CeO2 films with well-characterized architectures with 800 keV He + ions. Glancing incidence X-ray diffraction (GIXRD), Raman spectroscopy, and diffuse reflectance spectroscopy were used to measure the quantitative changes in structural correlations and the modification of electronic properties as a function of ion fluence. We show that the decrease of phonon lifetimes and the changes of the optical spectra depends on the excess polaron concentration produced by the irradiation. The implications of these findings for the properties and the radiation resistance of these systems are discussed
