Archive Ouverte INSA Lyon
Not a member yet
75807 research outputs found
Sort by
Deploying Compact and Reliable Deep Neural Networks in Safety-critical Applications
No full textInternational audienc
Sociologie de Saint-Étienne
No full textInternational audience"Tour à tour présentée comme une ville industrielle, une ville ouvrière, une " ville de foot " et, plus récemment, une ville en déclin, Saint-Étienne apparaît comme l'une des grandes perdantes des transformations de la société française. Ce constat n'a pas été démenti ces dernières années. Les problèmes économiques et sociaux ont même été redoublés par la crise sans précédent de la vie politique locale. Empêtrées dans les " affaires ", les élites politiques municipales ont poursuivi des stratégies inefficaces, aux effets inégalitaires, qui mêlent un soutien à l'attractivité d'espaces " vitrines " et la négligence des services essentiels à une vie digne pour tous.À l'heure où l'extrême droite prospère sur le discours des " fractures territoriales ", cet ouvrage aborde une réalité souvent occultée : celle des villes dont la situation échappe à la fois aux récits de la métropolisation vertueuse et à ceux qui la diabolisent. Il donne à voir une société urbaine plus contrastée que ne le suggèrent les discours misérabilistes sur la ville ou, au contraire, ceux qui célèbrent la " capitale du design ". Une société qui porte en elle des ressources dont certains habitants et collectifs se saisissent pour s'émanciper des logiques de compétition et de marchandisation.
Energy and Waste
No full textThe Anthropocene epoch is defined by the profound impacts of human activity on natural ecosystems. Industrial processing of raw materials produces vast quantities of man-made matter including materials, wastes, and residues, which Harpet (2001) termed the rudosphere. Monsaingeon (2017) further introduced the term poubellocene to characterize this era of waste and a wasteful world (UNEP 2024). In France, the article L541-1 from the Environment Code defines waste as “any residue from a production, transformation, or use process, any substance, material, product, or more generally any good or piece of furniture that is abandoned or that its holder intends to abandon.” In other words, the term waste is generic, qualifying any element that is abandoned
Depth-Resolved Analysis of Dimensional Stability (DS) in Poly(urethane-isocyanurate) Rigid Foams under Humid and Thermal Stress
No full textInternational audienceThis study examines the dimensional stability (DS) of polyisocyanurate rigid (PIR) foams under high relative humidity (RH = 90%) and elevated temperature (T = 70 °C), representative of the extreme conditions that may be encountered in roofing applications. Lateral expansion is observed in the plane perpendicular to the rise direction, which is strongly depth-dependent (Z-axis) and decreases from 5.3% for the surface layer (0−5 mm) to around 1.2% for the inner layers (Z ≥ 15 mm). This phenomenon is mainly attributed to the lower isocyanurate content near the surface and is amplified by the higher partial pressure of isopentane (+9%) near the surface. In contrast, contraction occurs in the rise direction (also called the thickness direction or TD) and remains nearly constant at −1% for Z ≥ 15 mm. This anisotropic behavior is attributed to the elongated cell morphology in the TD, inevitably leading to a decrease in mechanical strength in the plane perpendicular to the rise direction. Moisture content, quantified using Karl Fischer titration, is estimated at around 2.2 wt % (under humid conditions) and appears to play a dual effect, both plasticizing the polymer matrix and increasing internal gas pressure. Postcuring at T = 140 °C significantly increased crosslinking and improved DS, reducing foam expansion in the machine direction (MD) and cross-machine direction (CMD) from ΔL/L0 ≈ 5% to about 2% for the surface layer (0−5 mm). All these findings point toward the dominant role of moisture level, isocyanurate content, and cell shape factor of the final products on the long-term dimensional stability of PIR foams in humid, thermally stressed environments
Beyond Additive Design: An Empirical Taxonomy of Multimodal STEM Accessibility Systems
No full textInternational audienceMultimodal systems combining audio, haptic, and tactile channels are prevalent in STEM accessibility for blind and visually impaired users, yet real-world feedback reports high cognitive load despite technological advances. Through systematic analysis of 66 systems (2015--2025), we identify three architectural regimes: additive (channel stacking), augmentative (partial coordination), and integrative (orchestrated fusion). A five-dimensional scoring framework reveals dramatic architectural separation, yet conventional performance metrics show no regime variation---a decoupling we term the Differential Cognitive Yield (DCY) phenomenon: architecturally distinct systems impose dramatically different cognitive costs while yielding similar task performance. We contribute empirical design thresholds for perceptual integration and call for a shift from interface engineering to perceptual integration engineering
Digital instrument simulator platform to support the development of noninvasive optical NIR device for placenta monitoring
No full textInternational audienceSignificance: Abnormal placental development is a major cause of adverse pregnancy outcomes, but current methods for placenta monitoring are not suitable for bedside use. Continuous-wave near-infrared spectroscopy (CW-NIRS) is an optical technique that takes advantage of the near-infrared light to provide functional measurements such as tissue oxygenation at the bedside. However, the placenta is an organ located beneath several layers of tissue, making robust measurement of placental oxygenation with a CW-NIRS device a complex task.Aim: We propose a framework based on light propagation simulations to evaluate the sensitivity of CW-NIRS devices for placenta detection, along with tools to support NIRS instrument development for engineers. Approach:The maternal abdomen was modeled as a four-layer structure (i.e., skin, adipose tissue, muscle, and placenta). We used a numerical solution of the diffusion equation using a finite-element method to assess the sensitivity to measure placental function under various conditions (tissue layer thickness, skin tone, tissue oxygen saturation). We used a calibration procedure to evaluate the probability of acquiring a sufficient irradiation with a CW-NIRS device. We collected ultrasound abdomen images from 142 healthy pregnant participants that we segmented and digitized to demonstrate our approach.Results: With a Mini-CYRIL CW-NIRS device, we showed that placenta monitoring is not possible when using short integration time with a subject having a deep placenta (≥20 mm) and dark skin tones. With an integration time of 10 s and a temporal binning of 10 points, simulations indicated that subjects with very fair skin tone have a placenta-scanning probability of 12% at a placenta depth of 20 mm and 39% at a depth of 10 mm, using a 50 mm source-detector separation. Thick skin and dark skin tones act as a filter on the NIRS signal, blocking backscattered light and leading to greater absorption in deeper tissues. The spatially resolved spectroscopy method can be used to monitor placental oxygenation with a placenta close to the surface and an oxygen saturation in the muscle layer lower than that of the placenta. The simulation of a realistic cohort of 142 maternal abdomens aimed to identify the optimal acquisition conditions for CW-NIRS devices to be used in placental monitoring. Conclusions:We proposed a framework to evaluate and optimize CW-NIRS sensitivity for placenta detection. Further work is needed to improve the reliability of placental tissue oxygenation.</div
Path-conditioned training: a principled way to rescale ReLU neural networks
No full textDespite recent algorithmic advances, we still lack principled ways to leverage the well-documented rescaling symmetries in ReLU neural network parameters. While two properly rescaled weights implement the same function, the training dynamics can be dramatically different. To offer a fresh perspective on exploiting this phenomenon, we build on the recent path-lifting framework, which provides a compact factorization of ReLU networks. We introduce a geometrically motivated criterion to rescale neural network parameters which minimization leads to a conditioning strategy that aligns a kernel in the path-lifting space with a chosen reference. We derive an efficient algorithm to perform this alignment. In the context of random network initialization, we analyze how the architecture and the initialization scale jointly impact the output of the proposed method. Numerical experiments illustrate its potential to speed up training
Coping with unbearable heat. How and how much is air conditionning transforming cities and urban lives?
No full textInternational audienc
Segmentation et reconstruction 3D de lignées cellulaires par apprentissage profond à partir des images confocales
No full textNational audienceCe travail présente un pipeline complet combinant imagerie confocale, segmentation par deep learning et reconstruction 3D réaliste de cellules afin d’améliorer la modélisation radiobiologique en thérapies anticancéreuses utilisant des radioisotopes. En s’appuyant sur nnU-Net et sur une base de données soigneusement prétraitée, la segmentation du noyau et du cytoplasme atteint des performances élevées (Dice > 0,91). Les reconstructions 3D révèlent une forte hétérogénéité morphologique entre lignées cellulaires, incompatible avec des modèles géométriques simplifiés. La base de données obtenue constitue un support clé pour des simulations Monte Carlo avancées (Geant4-DNA), permettant d’affiner l’estimation du RBE et d’améliorer les modèles biophysiques tels que NanOx
Blockchain traceability valuation for perishable agricultural products: Balancing economic benefit and social impact
No full textInternational audienceThe adoption of blockchain-enabled traceability systems in agricultural supply chains offers farmers a means to reduce demand uncertainty. However, downstream retailers gain full visibility into product freshness, enabling selective purchases that may inadvertently increase food waste. This study evaluates the impact of such a traceability system by analyzing agricultural supply chain transactions under two scenarios: with and without blockchain implementation. By comparing order quantities and farmer profits in both cases, we find that blockchain adoption can enhance product sales because the smart contract effect. We also find that blockchain adoption can either amplify the bullwhip effect when circulation time is short or mitigate it when circulation time is long. The interaction between the bullwhip effect and the smart contract effect impacts the farmer’s profit. The farmer achieves higher profits using the blockchain-enabled traceability system if the smart contract effect outweighs the bullwhip effect. Furthermore, adoption costs play a crucial role in determining feasibility. Beyond economic implications, blockchain-enabled traceability systems also influence social outcomes, particularly in reducing food waste. Our analysis reveals four possible outcomes based on economic benefits and social impact: (i) win-win (higher profits and reduced waste), (ii) win-lose (higher profits but increased waste), (iii) lose-win (lower profits but reduced waste), and (iv) lose-lose (lower profits and increased waste). The likelihood of each outcome is strongly dependent on product shelf life—longer shelf life increases the probability of a win-lose scenario, while shorter shelf life raises the likelihood of a lose-win outcome. Win-win and lose-lose scenarios remain the least probable