HAL Portal IOGS (nstitut d'Optique Graduate School)
Not a member yet
12589 research outputs found
Sort by
On the theoretical limit of gradient descent for Simple Recurrent Neural Networks with finite precision
No full textInternational audienceDespite their great practical successes, the understanding of neural network behavior is still a topical research issue. In particular, the class of functions learnable in the context of a finite precision configuration is an open question. In this paper, we propose to study the limits of gradient descent when such a configuration is set for the class of Simple Recurrent Networks (SRNs). We exhibit conditions under which the gradient descent will provably fail. We also design a class of SRN based on Deterministic finite State Automata (DFA) that fulfills the failure requirements. The definition of this class is constructive: we propose an algorithm that, from any DFA, constructs an SRN that computes exactly the same function, a result of interest on its own
Lâchez les rênes, laissez-vous guider : installation facilitée et sécurisée d'eLabFTW
No full textInternational audienceThis poster entitled « Let go of the reins, let yourself be guided : easy and secure installation of eLabFTW » and based on an article was presented and discussed at the Journées Réseaux de l'Enseignement et de la Recherche (JRES) in Rennes in December 2024. A working group of ASR IT specialists from the RAISIN regional business network (Bordeaux) has produced this poster to offer you a summary of a support model for the deployment and configuration of eLabFTW software in a research unit, in line with the CNRS security guidelines.Ce poster intitulé « Lâchez les rênes, laissez vous guider : installation facilitée et sécurisée d'eLabFTW » et basé sur un article a été exposé et discuté aux Journées Réseaux de l'Enseignement et de la Recherche (JRES) de Rennes en décembre 2024. Un groupe de travail des informaticiens ASR du réseau de métier régional RAISIN (Bordeaux) a réalisé ce poster pour vous proposer un résumé d’un modèle d’accompagnement sur le déploiement et la configuration du logiciel eLabFTW dans une unité de recherche, et ce en adéquation avec les consignes de sécurité du CNRS
Capteur plasmonique diffracto-interférentiel basé sur un couplage de modes de plasmon excités par un réseau de diffraction
No full textInternational audienceL’effet de bascule énergétique dans un réseau métallique profond a permis de démontrer que deux ordres de diffraction pouvaient échanger de l’énergie via le couplage de modes de plasmon. Un tel effet est mis en œuvre pour le développement de nouveaux capteurs plasmoniques dont les performances atteignent des LOD de 10-7 RIU
Role of the Nottingham effect in heat transfer in the extreme near-field regime
No full textInternational audienc
Characterization of a perfect sinusoidal grating profile using an artificial neural network for plasmonic-based sensors
No full textInternational audienceIn this paper, we present a system intended to detect a targeted perfect sinusoidal profile of a diffraction grating during its manufactured process. Indeed, the sinusoidal nature of the periodic structure is essential to ensure optimal efficiency of specific applications as plasmonic sensors (surface plasmon resonance -based sensors). A neural network is implemented to characterize the geometrical shape of the structure under testing at the end of the laser interference lithography process. This decision tool operates in classifier mode prior to further processing. Then, the geometrical parameters of the structure can be reliably determined if necessary. Two solutions can be considered: the detection of a fixed geometrical shape operating on a binary mode and the identification of a geometrical shape from a limited number of profiles. These methods are validated in the context of plasmonic sensors on experimental sinusoidal grating structures with a grating period of 627 nm
Abrasive-free chemical-mechanical planarization (CMP) of gold for thin film nano-patterning
No full textInternational audienceThis 5 × 3 mm 2 photograph shows the vibrant colors reflected by a nanostructured gold surface made by CMP. Each 20 × 20 μm 2 pixel is a mosaic of gold nanocubes with different dimensions, creating this colorful image of the gold CMP damascene technique
Eco-friendly soft nanoimprinting of TiO2 nanostructures with a large range of pattern heights
No full textInternational audienceNanoimprint lithography (NIL) is nowadays the most popular and effective method to develop new environmentally-friendly and low-cost photonic nanodevices. Combined with titanium dioxide (TiO2) in the form of sol-gel, photonic nanostructures with low absorption and high refractive index can be produced, which can be of interest for many applications for which sustainability becomes increasingly important. In this paper, we present a patterning process based on soft NIL of TiO2 sol-gel, and show that the very good pattern transfer has been obtained, which can be completed with independence from the feature size, shape and height of the patterns. We also propose a low-temperature (400°C) calcination process rather than 450°C or higher temperature to crystallize the TiO2 nanostructures, which leads to very similar crystalline structures to higher-temperature processes, and a vertical shrinkage of about 61% compared to the imprinted pattern. Using this environmentally-friendly combined soft-NIL + calcination process, we show that submicron patterns with heights above 300 nm can be obtained. Such a large pattern height, combined with the wide range of pattern shapes and dimensions that can be fabricated, opens the possibility of a wide diversity of designs for the eco-friendly fabrication of TiO2 nanostructures with highly-interesting photonic properties