HAL-Université de Bretagne Occidentale
Not a member yet
75442 research outputs found
Sort by
A Blockchain-Enhanced Reversible Watermarking Framework for End-to-End Data Traceability in Federated Learning Systems
No full textInternational audienceIn federated learning (FL) environments, ensuring data traceability presents significant challenges, particularly when data move between multiple entities such as data centers, edge nodes, and data scientists. This paper presents a novel framework that combines robust reversible watermarking and blockchain technology to achieve end-to-end traceability of medical images in a FL context. Based on the watermark, it becomes possible to interrogate the blockchain about the life cycle of an image to ensure data traceability, authenticity, and integrity. We use a histogram shifting-based reversible watermarking scheme with a new overflow management procedure, integrated with a private blockchain that records all watermarking and verification operations. Experimental results demonstrate the effectiveness of our approach in terms of watermark robustness considering a chest X-ray image dataset. We further show that watermarking does not interfere in the training and inference phase of a VGG-16 classification model for a Covid-19 medical database. A model trained on protected data can be used to classify nonwatermarked data as well.</div
Rhematic Structure, a Squib
No full textInternational audienceBased on ideas derived from the Prague School, Halliday made a distinction between thematic structure and information structure. Thematic structure divides the clause into a theme and a rheme. While there is a vast literature on the nature of the theme, that on the rheme is sparse. At the same time the rheme is frequently long and complex, a fact which led some to suggest that the theme extends beyond the first major component, which Halliday had identified as constituting the theme. It is hypothesized that Halliday’s position is basically correct, but that where the rheme contains more than one component, these will be ordered by the speaker in decreasing order of thematic importance. This is illustrated with several authentic examples
Comparative Study of Memory Optimization Techniques for Dataflow-Modeled Applications
No full textInternational audienceEfficient memory management is essential for signal and image processing systems, particularly in data-intensive applications where performance and resource constraints are critical. This paper presents a comparative study of two advanced memory optimization techniques: Memory Script Optimization (MSO), and Passive Active Flow Graph (PAFG) Optimization-within the context of dataflow-modeled applications. Both approaches aim to reduce memory usage and improve execution efficiency, but they do so with distinct strategies: Memory Scripts focus on in-place buffer management, while PAFG modifies actor interactions to minimize buffer requirements. Using a portion of a Convolutional neural network (CNN) application as a case study, we evaluate the efficiency of these techniques in terms of memory reduction and execution time. Our results demonstrate that MSO provides significant performance improvements, achieving up to 17% memory savings and 21% faster execution times, making it ideal for independent data operations. However, PAFG offers greater scalability and flexibility, particularly when dealing with complex data dependencies, and provides a simpler path to implementation. This work not only highlights the tradeoffs between memory efficiency and flexibility but also paves the way for applying these optimizations in near-memory computing architectures, where distance from memory to processing is employed as a parameter to improve efficiency
Boundary-aware dynamic re-weighting for semi-supervised medial image segmentation
No full textInternational audienceDifferent from traditional semi-supervised learning (SSL), semi-supervised medical image segmentation faces two significant challenges: (1) the imbalanced distribution of labeled data causes models to bias towards majority classes; (2) the distribution discrepancy between labeled and unlabeled samples induces confirmation bias in pseudo-labels. Inspired by clinical practice, where experienced doctors utilize intrinsic features from the interior of target organs to clarify ambiguous boundaries and focus on minority classes, we propose a novel boundary-aware dynamic re-weighting network (BDRN). First, we utilize edge filters to generate visually different but semantically aligned views, compelling two sub-networks to learn informative features from organ interiors and boundaries, respectively. Second, we extract the boundary and interior regions using morphological operators and introduce a shape constraint to enhance feature learning. Additionally, a conflict-adversarial module promotes segmentation consistency between different views. Finally, we propose a dynamic re-weighting strategy based on the effective number to improve attention to imbalanced classes. Experiments demonstrate that our method significantly improves segmentation performance, achieving state-of-the-art results on CT and MR images. Ablation studies further confirm the efficacy of boundary consistency constraints and dynamic re-weighting. The segmentation Dice score for minority organs (e.g., esophagus) on the Synapse dataset is improved by 17.1 %, 46.2 %, and 49.4 % using 10 %, 20 %, and 40 % labeled data, respectively
A Retrospective on DISPEED - Leveraging Heterogeneity in a Drone Swarm for IDS Execution
No full textInternational audienceSwarms of drones are gaining more and more autonomy and efficiency during their missions. However, security threats can disrupt their missions' progression. To overcome this problem, Network Intrusion Detection Systems ((N)IDS) are promising solutions to detect malicious behavior on network traffic. However, modern NIDS rely on resource-hungry machine learning techniques, that can be difficult to deploy on a swarm of drones. The goal of the DISPEED project is to leverage the heterogeneity (execution platforms, memory) of the drones composing a swarm to deploy NIDS. It is decomposed in two phases: (1) a characterization phase that consists in characterizing various IDS implementations on diverse embedded platforms, and (2) an IDS implementation mapping phase that seeks to develop selection strategies to choose the most relevant NIDS depending on the context. On the one hand, the characterization phase allowed us to identify 36 relevant IDS implementations on three different embedded platforms: a Raspberry Pi 4B, a Jetson Xavier, and a Pynq-Z2. On the other hand, the IDS implementation mapping phase allowed us to design both standalone and distributed strategies to choose the best NIDSs to deploy depending on the context. The results of the project have led to three publications in international conferences, and one publication in a journal
Visco-elasto-plastic characterization and modeling of a wet polyamide laid-strand sub-rope for floating offshore wind turbine moorings
No full textInternational audiencePolyamide 6 fiber ropes are of interest for floating offshore wind turbine mooring lines but could exhibit complex mechanical behavior during loading at sea, such as creep, relaxation, variable dynamic stiffness or visco-plasticity. There is a need for a model that could be introduced into finite element analyses to predict this complex response; it should also describe the effect of the loading history. This paper presents a visco-elasto-plastic behavior model based on four dashpot-ratchet-spring elements that allow a precise description of polyamide 6 rope behavior. An identification method, using a multi-relaxation test, is described. It has been implemented in a finite element analysis software and validations are made by comparing the model results to the experimental data. The present work is the result of an extensive effort initiated by the collaborative research project POLYAMOOR and continued by the MONAMOOR project, both led by France Energies Marines
Expérience d’écoute d’une expérience maritime et politique
No full textInternational audiencePrésentation en collaboration avec le documentariste Thomas Bour d'un travail de mise en écoute de l'expérience maritime pendant une expédition au Groenland organisée par une association environnementaliste pour documenter les effets du changement climatique
Power Consumption Analysis for Reverse Engineering Digital Modulation: A Novel Approach to Physical Layer Attacks on Communication Systems
No full textInternational audiencePhysical layer security is a critical aspect of embedded wireless communication devices, particularly in military, civil, intelligence, and security applications. Ensuring secure communication at this layer is essential to prevent unauthorized access and exploitation of vulnerabilities. While many existing techniques focus on the transmitted signal, they often overlook potential side-channel vulnerabilities that arise when an attacker gains access to the device itself. In this work, we propose a novel approach to enhancing physical layer security by demonstrating the feasibility of differentiating modulation schemes through power consumption analysis (PCA). Using a Field Programmable Gate Array (FPGA), we implemented various modulation schemes and measured the associated power consumption traces. Based on these measurements, we developed a classification algorithm using the gradient-boosted decision tree method. Our approach achieved a classification accuracy of approximately 99%, highlighting the potential of power analysis as both a tool for identifying vulnerabilities and strengthening security in embedded wireless communication devices.</div
LIMIT THEOREMS AND LACK THEREOF FOR A MULTILAYER RANDOM WALK MIMICKING HUMAN MOBILITY
No full textWe introduce a continuous-time random walk model on an infinite multilayer structure inspired by transportation networks. Each layer is a copy of R^d , indexed by a non-negative integer. A walker moves within a layer by means of an inertial displacement whose speed is a deterministic function of the layer index and whose direction and duration are random, but with a timescale that depends on the layer. After each inertial displacement, the walker may randomly shift level, up or down, independently of its past. The multilayer structure is hierarchical, in the sense that the speed is a nondecreasing function of the layer index. Our primary focus is on the diffusive properties of the system. Under a natural condition on the parameters of the model, we establish a functional central limit theorem for the R d -coordinate of the process. By contrast, in a class of examples where this condition is violated, we are able to determine the correct scaling of the process while proving that no limit theorem holds