BSc / MSc Learning & plasticity
Credit assignment in spiking neural networks
Implement and compare biologically plausible credit-assignment rules against backpropagation-through-time on a small temporal classification task. Suitable for BSc or MSc.
The project
How does a network of spiking neurons learn when the error signal is delayed and sparse? In this project you will implement and compare biologically plausible credit-assignment rules (e-prop, surrogate gradients) against backpropagation-through-time on a small temporal classification task. A BSc scope benchmarks the existing rules; an MSc scope pushes into analysing why they diverge from the exact gradient. Either way you leave with a clean, reproducible comparison that feeds the lab's wider work on learning in recurrent spiking networks.
SupervisorPascal Nieters