LEAT

Offres

Design of antennas to be integrated into a wearable IoT device for gait analysis

The mission of PROLONG project is to monitor movements and activities of older adults in order to detect conditions of possible danger, such as falls, night wandering episodes, to name but a few.
The internship is aimed at designing multiple antennas to be integrated into the wearable PROLONG receiver. Antennas will be designed with the aim of maximizing the miniaturization while keeping an acceptable energy efficiency value.

Plus d'info

Antennes-Capteurs miniatures complexes réalisées en impression 3D

Dans ce stage, on s’intéressera dans un premier temps à démontrer le potentiel antennaire de la technologie d’impression 3D en ayant pour objectifs : (a) l’optimisation du rapport taille/efficacité/bande passante, et (b) Un meilleur volume d’occupation pour une efficacité maximale.
A partir de ces résultats, l’étude portera sur l’association des technologies additives et de métaux liquides afin de réaliser des structures antennaires 3D complexes.

Plus d'info

Fully dielectric reflectarrays made in 3D printing working in millimeter wave band

During this internship, we will first demonstrate the potential of realizing a reflectarray entirely in dielectric using 3D printing technology and operating in the millimeter range. Among the various points to be addressed, attention will be paid to the efficiency and bandwidth of the reflector array.
If the study gives encouraging results, the design of a dual band reflectarray will be considered.

Plus d'info

Etude dosimétrique (DAS) d’une tête soumise à une exposition dans la bande de fréquence de la 5G (3.5 GHz) dans le cadre d’un cumul d’exposition

Avec le déploiement de la 5G, la problématique des éventuels effets des ondes électromagnétiques sur la santé est plus que jamais d’actualité. Du fait de la difficulté des mesures, la simulation numérique est l’outil privilégié pour quantifier la puissance absorbée par les tissus (DAS, W/kg) et l’élévation de température qui en découle : on parle de dosimétrie numérique. Dans ce stage il s’agira de simuler, via le logiciel commercial CST, une tête hétérogène voxélisée soumise à 3.5 GHz puis à divers cumuls d’exposition. L’objectif sera donc d’obtenir des cartographies de DAS et de la distribution de température à 3.5 GHz puis dans le cadre d’un cumul d’exposition.

Plus d'info

Identification d’objets à partir de l’étude du champ diffracté : mise en évidence de l’intérêt du Machine Learning

Dans ce stage, on s’intéressera à l’utilisation de méthodes du type Machine Learning pour l’aide à la décision et l’identification d’objets. Ces traitements seront appliqués au champ diffracté par ces objets lorsqu’ils sont soumis à une onde électromagnétique Ultra Large Bande. Ces travaux de stage auront pour objectifs principaux : (a) l’enrichissement et la structuration de la base de données nécessaire aux méthodes d’apprentissage, (b) la sélection des algorithmes les plus pertinents, et (c) l’implémentation de ces méthodes en complément des méthodes déjà étudiées au laboratoire.

Plus d'info

Multimodal unsupervised learning with self-organizing maps

The international project SOMA (Self-Organizing Machine Architecture) aims at designing non-conventional computing architectures inspired by the plasticity of the brain.
In that context, we proposed at LEAT a computational model for brain-inspired multimodal association named ReSOM based on unsupervised learning with Self-Organizing Maps (SOMs) [2] and Hebbian-like learning [3]. The goal of this position is to lead the experimental study consisting of the deployment of the model onto FPGA-boards. The SCALP boards are developed by our partner HESSO in Geneva, Switzerland, to build modular computing structures [4]. Each neural map will be deployed on a specific board and then the maps will communicate with each other through high speed serial links to develop lateral synaptic connections. A quantitative study will be held to assess the performance of the system for multimodal unsupervised learning in terms of accuracy, dynamicity and scalability.

Plus d'info

Confrontation of supervised and unsupervised learning of Residual spiking neural networks

Spiking Neural Networks (SNNs) are considered as a new computing paradigm that can potentially replace existing solutions in the next generation of Artificial Neural Networks. In addition to the fact that SNNs are inherently more biologically plausible, they offer the prospect of event-driven hardware operation and energy efficient processing [1] .
Residual learning (RL) is part of the main recent achievements for training CNN models. However, these methods are still under exploration in the domain of SNN [3]. Indeed, one of the main challenge with SNN is to deal between efficient neural coding and good recognition performance whatever the depth of the network [2]. In this work, we consider the integration of such features in Deep Spiking networks to speed up the training process. This project takes place in the context of the Interdisciplinary Artificial Intelligence Institute 3IA Cote d’Azur in the field of bio-inspired AI.

Plus d'info

Deep spiking neural networks for embedded artificial intelligence

This subject takes place in a collaborative project with Renault to explore the use of SNN in automotive use cases. The goal is to bind those specific neural networks to Event-Based Sensors (EBS) for perception in self-driving applications. The association between SNN and EBS would enable to build a complete even-based processing chain [3, 4]. The goal of this internship is to develop a first software model of the processing chain. The internship will then continue during a PhD already funded by the ANR (Agence National de la Recherche).

Plus d'info

Toward on-chip learning of spiking neural networks from event-based sensors

The « neuromorphic » event-based approach to vision and image sensing is recently gaining substantial attention as it proposes solutions to the problems encountered with conventional technology of image processing. The output of such a sensor is a time-continuous stream of pixels, delivered at unprecedented temporal resolution, containing zero redundancy and encoding orders of magnitude higher dynamic range than conventional image sensors. However, due to the lack of alternatives so far, the event-based, asynchronous output of these sensors have been processed using conventional computing devices such as CPUs and GPUs. This way of processing is obviously non-ideal and does not allow to fully benefit from the unique characteristics of such sensors. In this postdoctoral project, we will attempt to develop an event-based processing chain applied to a realistic data in the context of moving objects. To reach this goal, we will explore and compare on-chip the most recent learning methods adapted to spiking neural networks.

Plus d'info