HAL Portal UPHF (Université Polytechnique Hauts-de-France)
Not a member yet
28473 research outputs found
Sort by
Event-Driven Neural Network for Ultra-Low-Power On-Chip Keyword Spotting
No full textInternational audienc
« Mais qu'est-ce qu'elle écrit, celle-là ? » : Les manuscrits de Marceline Desbordes-Valmore. Déchiffrages et incertitudes en paléographie.: Présentation du projet de transcription de la Société des Études Marceline Desbordes-Valmore
No full textInternational audienceAs part of the doctoral conference at Université Polytechnique Hauts-de-France (Valenciennes), devoted to uncertainty in research, Maxance Lardjane, co-organizer of the conference, presented Marceline Desbordes-Valmore's manuscripts.More specifically, he used the poet's manuscripts to discuss the difficulties of paleography.This subject also allowed him to introduce attestants to the website of the Société des Études Marceline Desbordes-Valmore, an association devoted to the author of Les Pleurs. He showed the public side of the website (browsing Desbordes-Valmore's poems in full text) as well as its research-only aspect (kilometer-long transcriptions of the poet's manuscripts, activities promoting Desbordes-Valmore's work, etc.).Dans le cadre des doctoriales de l'Université Polytechnique Hauts-de-France (Valenciennes) consacrées à l'incertitude en recherche, Maxance Lardjane, co-organisateur des doctoriales, a présenté les manuscrits de Marceline Desbordes-Valmore.Plus précisément, il s'est servi des manuscrits de la poétesse pour parler des difficultés inhérentes à la paléographie, partie indispensable de la philologie des auteurs anciens.Cela lui a aussi permis de présenter aux participants le site de la Société des Études Marceline Desbordes-Valmore, tant dans son côté public (navigation dans les poèmes desbordes-valmoriens en plein texte) que dans son côté réservé aux chercheurs (transcription au kilomètre des manuscrits de la poétesse, activités de valorisation de l’œuvre desbordes-valmorienne, etc).Cette intervention donnera lieu à des actes de colloque dans la (future) revue des doctorants de l'Université de Valenciennes
Un contrôle d'accès décentralisé efficace et finement granulaire pour les écosystèmes IoT sur NDN
No full textInternational audienceNamed Data Networking (NDN) is an emerging Information-Centric Networking (ICN) architecture that offers a fresh and dynamic approach to Internet communication. This architecture is well-suited for the IoT ecosystem due to its efficient data distribution strategies, which provide a secure and scalable data-centric environment. Nonetheless, NDN's philosophy results in less control over data access. To address this issue, several proposals in the literature tend to utilize encryption schemes. However, most of them fail to offer fine-grained access control for heterogeneous and multi-domain architecture. In addition, several proposals do not provide efficient revocation mechanisms that are suitable for scalable and resource-constrained environments. Moreover, the proposed encryption scheme do not usually accommodate NDN's and data caching and group communication features. Therefore, to overcome these challenges, we propose in this paper a new decentralized attribute-based access control scheme for IoT ecosystems over NDN. Our solution considers continuous data sharing scenario and ensures fine grained access to data. Furthermore, it enables a fast data access, through a secure key derivation mechanism, and an efficient and direct revocation. The encryption and decryption processes that follows the revocation are optimized in order to reduce the revocation overhead and maintain a fast access to data. Finally, our experimental results demonstrate that our solution outperforms existing ones.Le Named Data Networking (NDN) est une architecture émergente de réseau centré sur l'information (ICN) qui propose une approche nouvelle et dynamique de la communication sur Internet. Cette architecture est particulièrement adaptée à l'écosystème de l'Internet des objets (IoT) grâce à ses stratégies efficaces de distribution de données, qui offrent un environnement centré sur les données, à la fois sécurisé et évolutif. Cependant, la philosophie de NDN conduit à un contrôle limité sur l'accès aux données. Pour résoudre ce problème, plusieurs propositions dans la littérature ont recours à des schémas de chiffrement. Cependant, la plupart d'entre elles ne permettent pas un contrôle d'accès granulaire dans une architecture hétérogène et multi-domaine. En outre, plusieurs de ces propositions ne fournissent pas de mécanismes de révocation efficaces, adaptés à des environnements évolutifs et contraints en ressources. Par ailleurs, les schémas de chiffrement proposés ne tiennent généralement pas compte des caractéristiques de mise en cache des données et de communication de groupe propres à NDN. Ainsi, pour surmonter ces défis, nous proposons dans cet article un nouveau schéma de contrôle d'accès basé sur les attributs et décentralisé pour les écosystèmes IoT sur NDN. Notre solution prend en compte un scénario de partage continu des données et garantit un accès finement contrôlé aux données. De plus, elle permet un accès rapide aux données grâce à un mécanisme sécurisé de dérivation de clés et une révocation directe et efficace. Les processus de chiffrement et de déchiffrement qui suivent la révocation sont optimisés afin de réduire les surcoûts liés à la révocation et de maintenir un accès rapide aux données. Enfin, nos résultats expérimentaux démontrent que notre solution surpasse les solutions existantes
Conception bio-inspirée et technologie MEMS pour les robots volants miniatures
No full textNational audienceLa conception d’un robot volant miniature impose des contraintes strictes en matière de taille et de consommation énergétique, rendant inadaptées les approches classiques utilisées à plus grande échelle. Il est donc nécessaire de repenser entièrement chaque étape du développement. Ce travail propose une démarche bio-inspirée, fondée sur l'observation des insectes, ainsi que des concepts spécifiquement conçus pour relever ces défis. Une analyse approfondie, s’appuyant à la fois sur des simulations numériques et des essais expérimentaux, est ensuite présentée. Les prototypes réalisés, d’un poids de seulement 22 mg, avec une envergure de 25 mm et une longueur de 15 mm, sont capables de générer une force de portance équivalente à leur propre poids
High Round-Trip Gain Piezoelectric-Capacitive Hybrid Micromachined Ultrasonic Transducer Based on Anodic Bonding Technology
No full textInternational audienceIn this work, a piezoelectric-capacitive hybrid micromachined ultrasonic transducer (HMUT) with a high round-trip gain was developed to achieve superior pulse-echo imaging performance. Initially, an equivalent circuit model (ECM) and a finite element model (FEM) were constructed to design and simulate the HMUT cell. A patterned etching technique for the piezoelectric layer was incorporated into the HMUT design. Simulation results indicated that the round-trip gain of the HMUT with patterned piezoelectric layer is 33 dB higher than that of the piezoelectric micromachined ultrasonic transducer (PMUT) and 3 dB higher than that of the capacitive micromachined ultrasonic transducer (CMUT). One of the innovative aspects of this technique is its ability to fabricate vertically stacked PMUT and CMUT. Based on anodic bonding technology, prototypes of 3 MHz PMUT, CMUT, and HMUT were fabricated on the same substrate. Finally, an electric impedance test and pulse-echo test were conducted. Experimental data confirmed that the HMUT exhibits higher round-trip gain, consistent with the simulation results. The successful fabrication of these prototypes demonstrates the feasibility of integrating multiple ultrasonic transducers on a single platform, enhancing the device’s compactness and versatility. Furthermore, it opens new avenues for the development of high-quality ultrasonic imaging
Object detection based on Logistic Objects in Context (LOCO) dataset: an improved dataset split and performance on NVIDIA Jetson Nano
No full textInternational audienceThe Logistics Objects in Context (LOCO) dataset is the first and only public dataset that integrates multiple logistic objects in a real logistics context. It is dedicated to object detection applications. However, the dataset is small and has unbalanced classes annotations problem. In this study, we present significant advancements to use YOLO models on the LOCO dataset, emphasizing improved dataset management and optimized inference performance. We introduced a novel balanced dataset splitting, ensuring equitable annotation distribution, which is detailed through a comprehensive dataset analysis. Our results demonstrate enhanced accuracy for YOLOv4 and YOLOv8, while we present additional preliminary results for YOLOv9 and YOLOv10, enabled by the identified experimental parameters. Notably, we achieved state-of-the-art inference times for both YOLOv4-Tiny and YOLOv8n when deployed on the NVIDIA Jetson Nano. Additionally, we provide preliminary findings on accuracy and inference times for YOLOv9 and YOLOv10, revealing important performance overlaps among various small-scale YOLO models on the same platform. Finally, we propose the optimal model configuration for achieving superior real-time performance on the NVIDIA Jetson Nano using the LOCO dataset, contributing valuable insights for future developments in real-time logistics object detection
MAGNETIC BERNSTEIN INEQUALITIES AND SPECTRAL INEQUALITY ON THICK SETS FOR THE LANDAU OPERATOR
No full textWe prove a spectral inequality for the Landau operator. This means that for all f in the spectral subspace corresponding to energies up to E, the L 2 -integral over suitable S ⊂ R 2 can be lower bounded by an explicit constant times the L 2 -norm of f itself. We identify the class of all measurable sets S ⊂ R 2 for which such an inequality can hold, namely so-called thick or relatively dense sets, and deduce an asymptotically optimal expression for the constant in terms of the energy, the magnetic field strength and in terms of parameters determining the thick set S. Our proofs rely on so-called magnetic Bernstein inequalities.As a consequence, we obtain the first proof of null-controllability for the magnetic heat equation (with sharp bound on the control cost), and can relax assumptions in existing proofs of Anderson localization in the continuum alloy-type model.</div
PE-CLIP: A Parameter-Efficient Fine-Tuning of Vision Language Models for Dynamic Facial Expression Recognition
No full textVision-Language Models (VLMs) like CLIP offer promising solutions for Dynamic Facial Expression Recognition (DFER) but face challenges such as inefficient full fine-tuning, high complexity, and poor alignment between textual and visual representations. Additionally, existing methods struggle with ineffective temporal modeling. To address these issues, we propose PE-CLIP, a parameter-efficient fine-tuning (PEFT) framework that adapts CLIP for DFER while significantly reducing trainable parameters while maintaining high accuracy. PE-CLIP introduces two specialized adapters: a Temporal Dynamic Adapter (TDA) and a Shared Adapter (ShA). The TDA is a GRU-based module with dynamic scaling that captures sequential dependencies while emphasizing informative temporal features and suppressing irrelevant variations. The ShA is a lightweight adapter that refines representations within both textual and visual encoders, ensuring consistency and efficiency. Additionally, we integrate Multi-modal Prompt Learning (MaPLe), introducing learnable prompts for visual and action unit-based textual inputs, enhancing semantic alignment between modalities and enabling efficient CLIP adaptation for dynamic tasks. We evaluate PE-CLIP on two benchmark datasets, DFEW and FERV39K, achieving competitive performance compared to state-of-the-art methods while requiring fewer trainable parameters. By balancing efficiency and accuracy, PE-CLIP sets a new benchmark in resource-efficient DFER. The source code of the proposed PE-CLIP will be publicly available at https://github.com/Ibtissam-SAADI/PE-CLIP
