Spiking Neural Networks for event-based imaging and sensing (2021/22) (2022/23)

Spiking neural networks (SNNs) represent a new generation of artificial neural networks (ANN), particularly well adapted to event-like data, such as events recorded by neuromorphic cameras or asynchronous streams of particle detection events (photons, neutrons,…). In contrast to more classical ANNs, SNNs are able to capture efficiently complex spatio-temporal patterns by adopting neural structures mimicking biological brains. However, important challenges still need to be addressed to accelerate the deployment of such networks to different applications, including efficient network design and training and appropriate hardware for fast and low-consumption implementation. In this project, the student will investigate new statistical models and methods to design and train SNNs efficiently. A particular focus will be on probabilistic SNNs, which can benefit the Bayesian and variational inference formalisms for training. Possible applications investigated during the project include neuromorphic computing for computer vision tasks using event-cameras (for robotic and microscopy applications) and analysis of event streams for single-particle detectors such as arrays of single-photon avalanche diodes (SPADs), or neutron/gamma detectors.