HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Guiding waves through chaos: Universal bounds for targeted mode transport
No full textInternational audienceControlling wave propagation in complex environments is a central challenge across wireless communications, imaging, and acoustics, where multiple scattering and interference obscure direct transmission paths. Coherent wavefront shaping enables precise energy delivery but typically requires full knowledge of the medium. Here, we introduce a universal statistical framework for targeted mode transport (TMT) that circumvents this limitation and validate it on various platforms including microwave networks, two-dimensional chaotic cavities, and three-dimensional reverberation chambers. TMT quantifies the efficiency of transferring energy between specified input and output channels in multimode wave-chaotic systems. We develop a diagrammatic theory that predicts the eigenvalue distribution of the TMT operator and identifies the macroscopic parameters—coupling strength, absorption, and channel control—that govern performance. The theory provides explicit bounds for optimal TMT wavefronts and captures phenomena like statistical transmission gaps and reflectionless states. These findings establish design principles for energy delivery and information transfer in complex environments, with broad implications for adaptive signal processing and wave-based technologies
Unlocking 2D/3D+T myocardial mechanics from cine MRI: a mechanically regularized space-time finite element correlation framework
No full textInternational audienceAccurate and biomechanically consistent quantification of cardiac motion remains a major challenge in cine MRI analysis. While classical feature tracking and recent deep learning methods have improved frame wise strain estimation, they often lack biomechanical interpretability and tem poral coherence. In this study, we propose a spacetime regularized finite element digital image/volume correlation (FE DIC/DVC) framework that enables 2D 3D+T myocardial motion tracking and strain analysis using only routine cine MRI. The method unifies Mu ltiview alignment and 2D/3D+T motion estimation into a coherent pipeline, combining region specific biomechanical regularization with data driven based temporal decomposition to promote spatial fidelity and temporal consistency . A correlation based Multiv i ew alignment module further enhances anatomical consistency across short and long axis views. We evaluate the approach on one synthetic dataset (with ground truth motion and strain fields), three public datasets (with ground truth landmarks or myocardial masks), and a clinical dataset (with ground truth myocardial masks). 2D+T motion and strain are evaluated across all datasets, whereas Multiview alignment and 3D+T motion estimation is assessed only on the clinical dataset. Compared with two classical feat ure tracking methods and four state of the art deep learning baselines, the proposed method improves 2D+T motion and strain estimation accuracy as well as temporal consistency on the synthetic data, achieving a displacement RMSE of 0.35 pixel (vs. 0.73 pixel), an equivalent strain RMSE of 0.05 (vs. 0.097), and a temporal consistency of 0.97 (vs. 0.91). On public and clinical data, it achieves superior performance in terms of a landmark error of 1.96 mm (vs. 3.15 mm), a boundary tracking Dice of 0.80 0.87 (a 2 4% improvement over the best performing baseline), and overall registration quality that consistently ranks among the top two methods By leveraging only standard cine MRI, this work enables 2D/3D+T myocardial mechanics and provides a practical route toward 4D cardiac function assessment
Infinite and Transfinite Sequences for Continuous Petri Nets
No full textContinuous Petri nets (CPNs) form a model of (uncountably infinite) dynamic systems that has been successfully explored for modelling and theoretical purposes. Let the mode of a marking be the set of transitions fireable in the future. Along a firing sequence, the sequence of different modes is non increasing, and forms what we call the trajectory of the sequence. In CPNs, a marking can be reachable by a finite sequence, or lim-reachable by an infinite convergent sequence. The set of trajectories (resp. markings) obtained via lim-reachability (sometimes strictly) includes the set of trajectories (resp. markings) obtained via reachability. Here, we introduce transfinite firing sequences over countable ordinals and establish several results: (1) while trans-reachability is equivalent to lim-reachability, the set of trajectories associated with trans-reachability may be strictly larger than the one associated with lim-reachability; (2) w.r.t. trajectories, transfinite sequences over ordinals smaller than ω 2 are enough; and (3) checking whether a trajectory is achievable is NP-complete.We then turn to a more difficult problem: the specification, for all transfinite firing sequences, of their achievable signatures, i.e. the sequences of markings witnessing the changes of mode along the trajectory. In view of this goal, we define a finite symbolic reachability tree (SRT) that tracks the possible signatures of the system; in the SRT, a set of markings with same mode is associated with each vertex. We establish that, for bounded CPNs, reversibility holds inside the leaves of the SRT (which correspond to the long-run behaviours). From an algorithmic point of view, we show how to build an effective representation of the SRT in exponential time, even when the CPN is unbounded.The length of an infinite firing sequence, i.e. the cumulated amount of firings, may be finite or infinite. Contrary to finite length infinite firing sequences, those with an infinite length may not converge and in fact their study is more involved. First we introduce a notion of similarity and show that every infinite length convergent firing sequence has a similar finite length firing sequence. Then we provide a necessary and sufficient condition for the existence of an infinite length firing sequence and another one for the existence of a converging infinite length firing sequence yielding polynomial time algorithms for checking such an existence.</p
Emerging insights into the piezoelectric properties of cellulose: From macro to nanoscale structures
No full textInternational audienceNowadays, the increasing demand for sustainable energy has brought piezoelectric materials to the forefront due to their capability to convert mechanical energy into electrical energy. In response to increasing environmental concerns, cellulose has emerged as a promising piezoelectric material, owing to its availability, biocompatibility, sustainability, biodegradability and cost-effectiveness. Despite significant research on the use of various forms of cellulose for piezoelectric energy harvesting, a systematic review focusing on the factors that can affect the piezoelectric property in cellulose remains notably absent. The main goal of this review is to fill this gap by understanding the piezoelectric behaviour of cellulose at different hierarchical levels, from macro-scale natural materials to nano-scale structures. This review presents an overview of the general aspects of the piezoelectric effect, followed by a detailed examination of the piezoelectric properties of cellulose. It further explores the piezoelectric behaviour of cellulose-based natural materials. The review then addresses the piezoelectric characteristics of nanocellulose and regenerated cellulose in turn. Furthermore, the review examines cellulose-based hybrid materials and their piezoelectric properties. In conclusion, the review highlights the current challenges and outlines promising directions for future research in this emerging are
Evolutionary Pre-Prompt Optimization for Mathematical Reasoning
No full textInternational audienceRecent advancements have highlighted that large language models (LLMs), when given a small set of task-specific examples, demonstrate remarkable proficiency, a capability that extends to complex reasoning tasks. In particular, the combination of few-shot learning with the chain-of-thought (CoT) approach has been pivotal in steering models towards more logically consistent conclusions [Wei et al. 2022b]. This paper explores the optimization of example selection for designing effective CoT pre-prompts and shows that the choice of the optimization algorithm, typically in favor of comparison-based methods such as evolutionary computation, significantly enhances efficacy and feasibility. Specifically, thanks to a limited exploitative and overfitted optimization, Evolutionary Pre-Prompt Optimization (EPPO) brings an improvement over the naive few-shot approach, exceeding 10 absolute points in exact match scores on benchmark datasets such as GSM8k and MathQA. These gains are consistent across various contexts and are further amplified when integrated with self-consistency (SC)
Reframing Pattern: A Comprehensive Approach to a Composite Visual Variable
No full textInternational audienceWe present a new comprehensive theory for explaining, exploring, and using pattern as a visual variable in visualization. Although patterns have long been used for data encoding and continue to be valuable today, their conceptual foundations are precarious: the concepts and terminology used across the research literature and in practice are inconsistent, making it challenging to use patterns effectively and to conduct research to inform their use. To address this problem, we conduct a comprehensive cross-disciplinary literature review that clarifies ambiguities around the use of "pattern" and "texture". As a result, we offer a new consistent treatment of pattern as a composite visual variable composed of structured groups of graphic primitives that can serve as marks for encoding data individually and collectively. This new and widely applicable formulation opens a sizable design space for the visual variable pattern, which we formalize as a new system comprising three sets of variables: the spatial arrangement of primitives, the appearance relationships among primitives, and the retinal visual variables that characterize individual primitives. We show how our pattern system relates to existing visualization theory and highlight opportunities for visualization design. We further explore patterns based on complex spatial arrangements, demonstrating explanatory power and connecting our conceptualization to broader theory on maps and cartography
Adaptive LASSO Quantile Regression with Fixed Effects
No full textInternational audienceQuantile regression with fixed effects for longitudinal data accounts for individual-specific intercepts.When the number of individuals is large relative to the number of repeated measurements and the covariate dimension is high, we propose a dimension-reduction approach based on the least absolute shrinkage and selection operator (LASSO) penalty. Specifically, we extend adaptive LASSO quantile regression to accommodate longitudinal data. The proposed method retains oracle properties, including asymptotic normality and consistent variable selection. Monte Carlo simulations demonstrate that it performs best under moderate dimensionality, while also outperforming alternative methods in low-dimensional settings, though with smaller margins.The practical relevance of the approach is illustrated through two real-world applications: French electricity consumption and the Millennium Cohort Study. Using departmental-level data from France, we analyze non-residential electricity consumption to uncover patterns in energy demand. The log-transformed outcome remains markedly skewed, emphasizing the need for models capable of handling non-Gaussian distributions. Drawing on data from the Millennium Cohort Study, we examine factors associated with children's internalizing difficulties across different quantiles of the outcome distribution, identifying maternal mental health (Kessler scale) as the most influential predictor.</p
Corrosion fatigue mechanisms of high-strength steel wires in the presence of various aggressive ions
No full textInternational audienceThis work focuses on the corrosion fatigue of high-strength steel wires with diameters of 200 μm. The fatigue load consists of a bending rotation with a stress amplitude between 1300 MPa and 700 MPa. Various solutions at pH 6 containing either sulfate, a mixture of citrate and either phosphate or hydroxyl are selected to evaluate the role of each ion or their combinations in accelerating the breakdown of the wires. For the higher stress amplitude applied, the lifetime is almost independent of the solution, showing mechanically controlled damage. For the lowest stress amplitude, the corrosion-fatigue limit significantly evolves depending on the solution. Sulfate leads to the shortest lifetimes, whereas experiments performed in deionized water show the longest lifetimes. A combination of citrate with either phosphate or hydroxyl leads to intermediate lifetime. The effect of each ion on the stability of the iron hydroxide oxide and the iron dissolution mechanisms is discussed. The results highlight the role of the chemical affinity of these ions with the iron surface and their consequences in corrosion fatigue crack generation
Automated classification of subsurface impact damage in thermoplastic composites using depth-resolved terahertz imaging and deep learning
No full textInternational audienceReliable detection of barely visible impact damage is critical to ensure the structural integrity of composite components in service, particularly in safetycritical applications such as pressure vessels and transportation systems. This study presents a solution for detecting such damage in woven glass fiberreinforced thermoplastic composites using terahertz (THz) time-of-flight tomography and convolutional neural networks. THz provides non-contact, non-ionizing, high-axial-resolution imaging of subsurface and back-surface damage, addressing key limitations of surface-based inspection methods. While THz imaging alone may not always permit conclusive damage identification, we bridge this gap by training neural network classifiers on depthresolved THz B-scan images using ground truth from co-located X-ray microcomputed tomography. Among several pretrained architectures tested via transfer learning, DenseNet-121 exhibits the highest accuracy. The model remains robust even when trained on truncated B-scans excluding surface indentation features, confirming its ability to detect structural anomalies located internally or on the back surface. This is particularly relevant for applications where back-side access is not feasible. Experimental validation is performed on impacted glass-fiber-reinforced thermoplastic coupons prepared in accordance with ASTM D7136, with damage severity quantified through force-displacement data and micro-tomographic analysis. Labeling for supervised learning conforms to acceptance criteria from industrial standards for composite pressure vessels (ASME BPVC Section X, CGA C-6.2), ensuring regulatory alignment and enabling deployment in quality control workflows. The proposed method minimizes the need for expert interpretation or secondary validation and offers direct applicability to in-service inspection and manufacturing quality control
Does data augmentation help or hinder the generalization of deepfake video detection?
No full textInternational audienceDeepfake detection models often fail to generalize across unseen manipulations and real-world video degradations. Despite numerous proposed methods, prior work lacks a systematic assessment of how data augmentation strategies affect forensic robustness. In this study, we perform a comprehensive evaluation of 14 data augmentation techniques for deepfake detection. Using the Xception backbone across FaceForensics++, DFDC-P, and Celeb-DF, we show that data augmentation improves both in-domain accuracy and cross-dataset generalization. In particular, frequency-aware strategies, such as FourierMix, substantially improve accuracy by up to +2–3% across multiple forgery types by amplifying spectral inconsistencies introduced during synthesis, while JPEG compression models real-world degradations and enhances decision confidence. Despite these gains, general-purpose, augmentation-enhanced models remain outperformed by specialized forensic architectures, such as Face X-ray and LipForensics by more than 10%, which explicitly leverage manipulation-specific cues. These findings provide practical guidance for designing model-agnostic augmentation pipelines and highlight the importance of hybrid approaches that combine effective data augmentations with specialized forensic architectures to achieve reliable and robust deepfake detection. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026