The proliferation of smart edge devices, from wearables to autonomous systems, creates a growing demand for local intelligence under tight energy constraints. Neuromorphic computing, taking inspiration from the remarkable efficiency of the brain, offers a promising answer, leveraging event-driven, spike-based computation to achieve energy-efficient processing at the edge.
This seminar builds on the neuromorphic foundations previously introduced and focuses on a central open challenge: on-chip learning. While classical deep learning relies on backpropagation, this algorithm is poorly suited to neuromorphic hardware due to its high memory and compute requirements. The seminar will survey what learning capabilities exist in today's neuromorphic chips, introduce bio-plausible local learning rules as a hardware-friendly alternative, and present ongoing research into making on-chip learning work without backpropagation.
5/6/2026 11:00 - 12:00
ESAT Aula L
