Description
Date depot: 27 novembre 2022
Titre: Federated learning for spiking neural networks with respect to neuromorphic solutions for energy-efficient hardware information processing
Directeur de thèse:
Farid NAIT-ABDESSELAM (LIPADE)
Directeur de thèse:
Aziz BENLARBI-DELAÏ (GeePs (EDITE))
Encadrant :
Siqi WANG (GeePs (EDITE))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Systèmes et architectures intégrés matériel-logiciel
Resumé: The end of Moore’s Law observed since 2005 brought back the interest to the neuromorphic approach, whose first work started in the early 40’s. This paved the way to the concept of bio-inspired and biomimetic engineering, consisting of new electronic solutions based on the observation of living things in order to analyze, in a short time and with extremely low energy consumption, complex and dynamic situations.
The spiking neural network (SNN) is one of the technological solutions inspired by the functioning of the brain. Similar to the cerebral cortex, neurons in the SNN are excited by a current which results in a spike-shaped membrane potential transmitted from one neuron to another through synapses. Not like neurons in ANN which are always kept active for data processing and memory access, a neuron in SNN shall be active only when it fires a spike. The SNN is believed to have over 100 times higher efficiency on energy consumption than the traditional ANN when implemented on a field-programmable gate array (FPGA).
Federated learning enables energy efficiency from the algorithmic aspect by training the neural networks across multiple decentralized edge devices or servers, holding local data samples, without exchanging them. An SNN trained by a federated learning algorithm will have lower energy consumption from both software and hardware perspectives.
This project will fill the gap between the two aspects of electronic and computing. The mathematical models, constructed according to the layout of spiking e-neuron circuits, will be taken as the base in the SNN computation.
Doctorant.e: Wei Yongtao