Reproduce “EchoSpike Predictive Plasticity: An Online Local Learning Rule for 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

Lars Graf, Zhe Su, Giacomo Indiveri

EchoSpike Predictive Plasticity (ESPP), developed by researchers at ETH Zurich, presents an online local learning rule for spiking neural networks, utilizing a self-supervised predictive plasticity mechanism based on past activity "echoes". It achieves competitive performance on neuromorphic datasets, demonstrating scalability to deep SNNs and suitability for resource-constrained neuromorphic hardware.

Key insights

  • Utilizing the accumulated spiking activity of the entire previous sample (the 'echo') for prediction within the current sample simplifies the learning rule and conserves resources for neuromorphic implementation.
  • The adaptive threshold and event-based gating mechanism within ESPP intrinsically regulate spiking activity and select the most relevant time steps for learning, leading to efficient, sparse weight updates.
  • ESPP's layer-wise objective functions enable effective scaling to deep SNN architectures and allow all network layers to contribute to downstream classification tasks, rather than relying solely on the final layer.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.