Training Neuromorphic Devices with Spike-Based Feedback Control Algorithm - Applications for Robotic Control

This project aims to develop a novel framework for training Spiking Neural Networks (SNNs) on neuromorphic devices (DYNAP-SE) using spike-based local learning and control feedback, enabling online learning without back-propagating gradients.

Running Master's Thesis

Description

This project addresses the challenge of training multi-layered Spiking Neural Networks (SNNs) for robotic control. For this purpose, a novel learning framework will be implemented on mixed-signals neuromorphic devices (the DYNAP-SE). The approach combines spike-based local learning and control feedback signals and thus does not require backpropagating gradients, allowing for online learning in SNNs on neuromorphic devices. After mapping the model onto the DYNAP-SE chip, the goal is to evaluate its performance on a control problem in a virtual environment using a computer-in-the-loop setup.

Supervisors

To the top of the page