Archive Ouverte INSA Lyon
Not a member yet
75807 research outputs found
Sort by
Optimization of structural and poling strategies in piezoelectric elastomer composites for soft sensing applications
No full textInternational audiencePiezoelectric flexible sensors are emerging as key components in medical applications, offering unique electromechanical properties for various diagnostic and therapeutic purposes. In this study, ceramic-filled silicone composites were developed as high-performance piezoelectric materials suitable for soft biomedical sensing applications. To enhance their electromechanical response, a multi-parametric design strategy was adopted, combining three approaches: the use of bimodal particle size distribution and the dielectrophoretic alignment of these particles within the matrix, supported by an optimized poling process. Results revealed that composites with an oriented particle distribution, consisting of 25 % of micro-sized particles and 75 % of nano-sized particles exhibited significant improvements in piezoelectric coefficient (d 33 ) compared to composites with randomly distributed particles. Additionally, the piezoelectric transverse coefficient (d 31 ) was significantly improved under in situ poling conditions, particularly in nano-rich and hybrid systems. These findings underline the potential of combining particle alignment, size hybridization, and poling optimization in enhancing the performance of piezoelectric composites for innovative medical sensor applications
Generalization and Scaling Laws for Mixture-of-Experts Transformers
No full textWe develop a theory of generalization and scaling for Mixture-of-Experts (MoE) Transformers that cleanly separates active per-input capacity from routing combinatorics. Conditioning on fixed routing patterns and union-bounding across them, we obtain a sup-norm coveringnumber bound whose metric entropy scales with the active parameter budget and incurs a MoE-specific overhead. Combining this with a standard ERM argument for squared loss we provided a generalization bound under a d-dimensional manifold model (d is the intrinsic dimension of the training data) and C β targets, showing that approximation and estimation trade off in the same way as in dense networks once active parameters are counted appropriately. We further prove a constructive approximation theorem for MoE architectures, demonstrating that accuracy can be improved either by scaling active capacity or by increasing the number of available experts, with the better of the two mechanisms prevailing. From these results we derive neural scaling laws, covering model scaling, data scaling and compute-optimal tradeoffs. The theory highlights that enlarging the expert pool at fixed sparsity influences performance only through a mild logarithmic routing term, whereas increasing active capacity per input drives the main gains in generalization and approximation. These insights provide principled guidance for the design of efficient sparse Transformer systems and clarify the fundamental tradeoffs underlying their empirical scaling behavior.</div
Du serment doctoral d'intégrité scientifique à un serment personnel : un atelier d'écriture et de réflexion sur la responsabilité et le rôle des scientifiques dans la société
No full textNous présentons un atelier de réflexion sur le serment doctoral d’intégrité scientifique et d’écriture d’un serment personnel, destiné aux doctorantes et doctorants, et plus généralement au personnel de la recherche. L’atelier est proposé depuis 2025 comme formation à l’éthique de la recherche dans quelques écoles doctorales en France. Avec un dispositif original, il permet d’examiner plusieurs aspects de la pratique et des enjeux sociaux-environnementaux de le Recherche : la responsabilité des scientifiques, l’engagement, le rôle des sciences dans l’anthropocène, la place de l’éthique et de l’intégrité dans la pratique du doctorat et des sciences en général
Development of an experimental method for the characterization of kinetic constants of hydrogenotrophic methanogenesis in mixed cultures: Overcoming mass-transfer limitations
No full textInternational audienceHydrogenotrophic methanogenesis is a key reaction in biological methanation processes. This reaction is known to be driven by both dynamics: the biological reaction itself, and the gas-to liquid mass transfer. Consequently, identifying biokinetic parameters of hydrogenotrophic methanogenesis is challenged by gas-liquid mass transfer limitations. To overcome this issue, an experimental setup was designed to operate at high pressure. Working at a high H2 partial pressure and at low biomass concentration enabled biologically driven dynamics to be observed, followed by mass transfer limitation. The transition towards the mass transfer regime was characterized by analyzing pressure dynamics. A numerical model was used to confirm this transition and to identify the biological kinetic parameters. The maximal growth rate in thermophilic mixed culture was between 0.18 and 0.22 h - 1
Comparison of non-Newtonian models in a bearing with a porous layer: Viscoelastic (Maxwell) versus micropolar (couple stress)
No full textThe tribology literature abounds with articles concerning the effects of non-Newtonian lubricant models on bearing behavior. Different categorizations are possible, but we look at a viscoelastic model, and a micropolar model. The bearing we consider contains a porous layer with flow properties described by the Darcy model. The two continuum models: the Upper Convected Maxwell model (UCM) and the Stokes Couple Stress model (CS) are based on entirely different underlying physical assumptions. In the UCM model, the fluid is characterized by a time scale, namely, the relaxation time. In the CS model, the fluid is characterized by a length scale, representing the microstructure size. However, when the thin film assumptions are applied, the governing equations of the two approaches look surprisingly similar. According to computed results, viscoelasticity tends to increase the pressure. This effect is far more pronounced in the steep inclination case. The couple stress length parameter likewise tends to increase pressure, at both moderate and steep inclination. In all cases the porosity tends to decrease the pressure, due to an effective softening of the confining surface
Reducing the Hardware Gap for Custom Accelerators through Quantization Aware Training
No full textInternational audiencePopular machine learning frameworks like PyTorch and TensorFlow provide complete toolchains to design, train, quantize and deploy convolutional neural networks (CNNs) on standard hardware (CPUs, GPUs, TPUs, NPUs, microcontrollers). However, custom hardware targets like FPGAs, ASICs and research accelerators typically use non-standard number representations as well as limited operator sets, and are poorly served by current lowering backends. Existing backend-driven approaches rigidly force training to their internal formats and often hide lowering choices behind opaque layers and closed software development kits. Conversely, unconstrained quantization that permits arbitrary numeric and operator freedom often produces models that are difficult to implement efficiently, leading to an hardware gap between the trained and deployed model. This work introduces HATorch, a PyTorch-based hardwareaware training framework that reduces this gap by enhancing quantization with hardware architecture model on the arithmetic operator/format level. HATorch supports custom hardware-friendly quantization flows, exposing lowering decisions for transparent model-hardware co-design
Multiresolution Adaptive Block-Coordinate Forward-Backward for Image Reconstruction
No full textClassical first-order optimization methods for imaging inverse problems scale poorly with image resolution. Wavelet based multilevel strategies can accelerate convergence under strong blur, but their fixed coarse-to-fine schedules lose effectiveness in moderate-blur or noise-dominated regimes.In this work, we propose an adaptive multiresolution block coordinate Forward-Backward algorithm for image restoration. Multiresolution block selection is driven by the local magnitude of the proximal update via a stochastic non-smooth Gauss-Southwell rule applied to the wavelet decomposition of the image. This adaptive selection strategy dynamically balances updates across scales, emphasizing coarse or fine blocks according to the degradation regime. As a result, the proposed method automatically adapts to varying blur and noise levels without relying on a predefined hierarchical update scheme
Past2ECO : Des cultures du passé aux cultures du futur : utilisation de solutions numériques pour mobiliser la diversité génétique adaptative pour l'agroécologie.
No full textNational audienc
An Analysis of Client-and Server-Side Google Tag Manager and its Tags on the Web
No full textInternational audienceGoogle Tag Manager (GTM) enables website Publishers to install analytics, advertising, and other tracking services as "Tags" on their websites. Traditionally, in Client-Side GTM, Tags collect and send data directly from the browser to third-party servers. However, as today's browsers and extensions increasingly block third-party trackers, Google introduced a Server-Side GTM in 2020. This GTM version allows to install and execute Tags directly on a first-party server, thus hiding the presence of third-party trackers. In this study, we perform a crawl of 80K popular websites worldwide to analyze the adoption of Client-and Server-Side GTM and the prevalence of their Tags. Our results show that GTM is present on 28.8% of websites, with 6.7% of these sites obfuscating its presence. Our Tags detection methodology is able to measure the prevalence of all GTM Tags -Official, Template, HTML and Image Tags -globally and reveals a strong dominance of Tags provided by Google, which appear on 95.3% of sites. We further analyze 386K HTML Tags and find that they invoke third-party libraries on at least 81.2% of websites, to potentially share personal data. Finally, we detect the presence of Server-Side GTM on 3K websites and 398 Server-Side Tags instances on the Web. We also propose a GTM-Eye browser extension that detects GTM and its Tags accessible to everyone