HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
Inverse Design of Mirror‐Symmetric Disordered Systems for Broadband Perfect Transmission
No full textInternational audienceWe present a framework for achieving broadband perfect wave transmission in complex systems by optimizing symmetric disordered media via inverse design. We show that leveraging symmetry of complex media reduces the optimization's complexity enabling the incorporation of additional constraints in the parameter space. Starting from a single perfectly transmitting state with predefined input and output wavefronts at a specific frequency, we progressively broaden the bandwidth — from a reflectionless exceptional point with a flattened lineshape to narrowband filters and ultimately to broadband quasi‐perfect transmission exhibiting a rainbow effect. Numerical simulations based on the coupled dipole approximation are validated experimentally in a multichannel microwave waveguide with dielectric and metallic scatterers. Finally, we demonstrate broadband enhanced wave transmission through barriers highlighting the potential for advanced wave control applications
Dynamic Agent Generation for Self-Adaptive Root Cause Analysis
No full textInternational audienceModern software systems increasingly rely on microservice architectures, which enhance modularity and resilience but produce vast amounts of heterogeneous observability data—logs, metrics, and traces—to ensure reliable operation and early failure detection. Performing Root Cause Analysis (RCA) on such data is challenging due to its scale, heterogeneity, and evolving structure, which hinder effective correlation and reasoning across modalities. Although recent studies have explored statistical techniques, graph-based models, and Large Language Models (LLMs)-based agents for RCA, most remain static and task-specific, lacking the adaptability and coordination required to handle evolving diagnostic contexts. This paper introduces a self-adaptive agent generation framework that leverages LLMs to dynamically compose and orchestrate adapted diagnostic agents at runtime according to each anomaly’s characteristics. Two main agents drive this process: a Parser Agent that interprets natural-language queries and builds structured task specifications, and an Executor Agent that adapts and coordinates adapted agents analyzing multimodal observability data through a shared memory space. Experiments on Nezha (fault-injected) and OpenRCA (real-world) datasets show up to 12% and 22% gains in diagnostic accuracy, confirming the framework’s effectiveness in adaptive reasoning, coordination, and interpretable root cause identification
Modeling Sampling Workflows for Code Repositories
No full textInternational audienceEmpirical software engineering research often depends on datasets of code repository artifacts, where sampling strategies are employed to enable large-scale analyses. The design and evaluation of these strategies are critical, as they directly influence the generalizability of research findings. However, sampling remains an underestimated aspect in software engineering research: we identify two main challenges related to (1) the design and representativeness of sampling approaches, and (2) the ability to reason about the implications of sampling decisions on generalizability. To address these challenges, we propose a Domain-Specific Language (DSL) to explicitly describe complex sampling strategies through composable sampling operators. This formalism supports both the specification and the reasoning about the generalizability of results based on the applied sampling strategies. We implement the DSL as a Python-based fluent API, and demonstrate how it facilitates representativeness reasoning using statistical indicators extracted from sampling workflows. We validate our approach through a case study of MSR papers involving code repository sampling. Our results show that the DSL can model the sampling strategies reported in recent literature. CCS Concepts• Software and its engineering → Domain specific languages.</div
Unsupervised Detection of Post-Stroke Brain Abnormalities
No full textInternational audiencePost-stroke MRI not only delineates focal lesions but also reveals secondary structural changes, such as atrophy and ventricular enlargement. These abnormalities, increasingly recognised as imaging biomarkers of recovery and outcome, remain poorly captured by supervised segmentation methods. We evaluate REFLECT, a flow-based generative model, for unsupervised detection of both focal and non-lesional abnormalities in post-stroke patients. Using dual-expert central-slice annotations on ATLAS data, performance was assessed at the object level with Free-Response ROC analysis for anomaly maps. Two models were trained on lesion-free slices from stroke patients (ATLAS) and on healthy controls (IXI) to test the effect of training data. On ATLAS test subjects, the IXI-trained model achieved higher lesion segmentation (Dice = 0.37 vs 0.27) and improved sensitivity to non-lesional abnormalities (FROC = 0.62 vs 0.43). Training on fully healthy anatomy improves the modelling of normal variability, enabling broader and more reliable detection of structural abnormalities
Doing Attending in Multi-Party Dinner Settings : Static and Dynamic Forms of Attention in French and French Sign Language
No full textInternational audienceIn social interaction research, so-called "listeners" are known for being active co-participants of the interaction through several engagement displays, labeled as feedback, backchannel, or listener responses. Enriched by our account of interactions in French and French Sign Language, we suggest using the term 'doing attending' so as to not restrict this practice to a single modality and highlight its functional and interactional nature. Our analyses of video-recorded interactions during family dinners held at home, further demonstrate how such multimodal displays may not always be characterized by 'dynamic' forms, and are deeply shaped by polyadicity as well as co-activity and material affordances, in both languages
Assessing Product Reliability from Customer Reviews Through Natural Language Processing and Machine Learning
No full textProduct reliability is important for both companies and consumers as it directly determines the market competitiveness of the product. Traditionally, to assess product reliability, companies need to collect large amount of time-to-failure data, either from in-lab testing or from the field. As reliability of modern products grows, it is becoming increasingly difficult to collect enough time-to-failure data to support traditional reliability assessment approaches. Meanwhile, with the growth of the Internet, many customers share feedback on the purchased products by posting online reviews. The posted reviews constitute a huge, easily accessible, and more realistic database that can be used to assess product reliability. How to extract information from these review data and use it to support reliability assessment, is, however, a relatively unexplored area. In this work, a dataset for detecting failure from customer review data was created, along with an evaluation baseline created by multiple human annotators to consider the ambiguity in natural language. Then, a systematic framework is developed for assessing product reliability from customer reviews. In the developed framework, failure information is extracted from the original data through Natural Language Processing (NLP) and machine learning, and used to support reliability assessment. Two NLP models are developed to detect failure occurrence and assess its severity from customer reviews. The first model relies on traditional NLP techniques like termweighting vectorization to create embeddings from raw texts. An ensemble learning model is then developed for failure detection. Our experiment shows that the best traditional NLP-based model can achieve 83.9% balanced accuracy and 79.2% F1-score for failure detection. However, compared to human performance (91.2% balanced accuracy and 89.6% F1-score), there is still a significant gap. In the second model, we utilized various transformer-based models and fine-tune them to detect the failure. The best-performing model, a fine-tuned DeBERTa-v3 base, achieved 88.5% balanced accuracy and 86.0% F1-score for failure detection, approaching the human performance. The best NLP model was finally applied on a real-world dataset to generate lifetime data for the products and assess their reliability and mean-time-to-failure. Overall, this research provides a first end-to-end methodology for assessing product reliability of products from online customer reviews and enables a comparison of product reliability reputations from real-world feedback. Code and data are available here: https://github.com/jmpion/CReFaDet
Shank motion analysis for quantifying knee gait deviations: Normative data at various walking speeds
No full textInternational audienceBackground: Knee deviations in clinical gait analysis are often measured using joint angles and could be enhanced by incorporating additional indices such as shank angular velocity in addition to knee angular velocity, particularly in the context of knee extensor thrust. The aim of this study was to establish normative data for shank and knee angular velocities at various gait speeds and to perform clinical validation of these biomarkers.Methods: A public dataset containing three-dimensional motion capture established on 50 healthy participants was used to calculate normative data of knee and shank angular velocities at five gait speeds (means, standard deviations and confidence intervals). Eleven hemiparetic persons walking with knee extensor thrust underwent three-dimensional gait analysis, during which minimum knee and shank angular velocities during stance phase were calculated.Results: Hemiparetic persons walking with knee extensor thrust had significantly lower minimum knee and shank angular velocities than normative data at all gait speeds (P < 0.001). In healthy persons, the minimum values for knee and shank angular velocities during stance phase or the values at foot-off, like most knee kinematics parameters, were correlated with gait speed (r = -0.83; r = 0.67; r ≥ 0.9; P < 0.01).Conclusion: This study provides normative data for knee and shank angular velocities at various gait speeds and demonstrates their usefulness for analysing knee deviations, notably the knee extensor thrust. This study underscores the impact of walking speed on gait patterns particularly on knee and shank angular velocities and underlines the need to compare data at the same gait speed for practice and future research
Quantitative Analysis of X-ray angiography images in Acute Ischemic Stroke
No full textNational audienceIschemic stroke is the most common type of stroke and is a major cause of death and disability worldwide. It occurs upon the obstruction of a cerebral artery by a blood clot, which causes the interruption of blood flow in the downstream brain territory. However, the reopening of the artery does not necessarily entail the restoration of complete reperfusion of the brain. CT or MRI perfusion imaging enable the assessment of brain perfusion but they cannot be used intra-operatively and are very seldom used for post-treatment follow-up due to clinical constraints. Means of a reliable and quantitative perfusion assessment would be a major asset if available during the intervention. The objective of this work is to generate brain perfusion maps based on intra-operative Digital Subtraction Angiography (DSA) image sequences
Programmable wave-domain computing in wireless communications
No full textFuture wireless networks must achieve large gains in data rate, energy efficiency and latency while integrating sensing and computation. Programmable wave-domain computing (pWDC), which processes signals directly through reconfigurable wavematter interactions, offers a way to offload part of this burden from electronic processors. In this Perspective, we review historical roots and recent work on pWDC and discuss its promise and challenges. We focus on three challenges: prototype-aware runtime optimization of pWDC hardware, enriching functionalities beyond linear continuouswave operation, and integrating pWDC into network-level resource management. Finally, we outline open questions regarding expressivity, practicality, and security that will arise in the transition of pWDC to real-life deployment in future wireless infrastructures
FSMODNet: A Closer Look at Few-Shot Detection in Multispectral Data
No full textFew-shot multispectral object detection (FSMOD) addresses the challenge of detecting objects across visible and thermal modalities with minimal annotated data. In this paper, we explore this complex task and introduce a framework named "FSMODNet" that leverages cross-modality feature integration to improve detection performance even with limited labels. By effectively combining the unique strengths of visible and thermal imagery using deformable attention, the proposed method demonstrates robust adaptability in complex illumination and environmental conditions. Experimental results on two public datasets show effective object detection performance in challenging low-data regimes, outperforming several baselines we established from state-of-the-art models. All code, models, and experimental data splits can be found at https://anonymous.4open.science/r/Test-B48D