Reproduce “SuperSpike: Supervised learning in multi-layer spiking neural networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Friedemann Zenke, Surya Ganguli
Researchers introduced SuperSpike, a deterministic, voltage-based three-factor learning rule designed to train multi-layer spiking neural networks (SNNs) for precise spatiotemporal spike pattern generation and non-linear classification tasks. The study also compared different credit assignment strategies, demonstrating their task-dependent efficacy.
Key insights
- SuperSpike's voltage-based, nonlinear Hebbian term and synaptic eligibility traces enable multi-layer SNNs to learn complex, nonlinear transformations of precise spatiotemporal spike patterns from a quiescent state.
- The effectiveness of credit assignment strategies is task-dependent; while simpler tasks can be solved with symmetric, random, or uniform feedback, complex spatiotemporal tasks require more structured feedback, like symmetric weights.
- The nonlinearity inherent in the SuperSpike rule, specifically the voltage-dependent term, is critical for achieving high performance on complex SNN learning tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.