Helping Spiking Neural Networks Can Learn to Learn

What this episode covers

Professor Emre Neftci, director of the Neuromorphic Software Ecosystems group at the Peter Grünberg Institute (PGI) within the Jülich Research Centre, explains how surrogate gradient networks make backpropagation compatible with spiking neural networks. He details how 'learning to learn' algorithms address device variability in emerging hardware like memristors and solve catastrophic forgetting in online systems. The discussion highlights applications in low-latency robotics, brain-machine interfaces for prosthetics, and energy-efficient smart city vision processing.