Reproduce “Brain-Inspired Learning on Neuromorphic Substrates” 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, Emre O. Neftci

Researchers introduce a mathematical framework for developing online learning algorithms on neuromorphic substrates, reducing the computational and memory complexity of Real-Time Recurrent Learning (RTRL) for Spiking Neural Networks. This approach enables gradient-based optimization while adhering to hardware constraints and demonstrates competitive performance on classification tasks.

Key insights

  • Real-Time Recurrent Learning (RTRL) can serve as a normative mathematical foundation for designing biologically plausible, local learning rules for Spiking Neural Networks (SNNs).
  • Approximating RTRL by ignoring explicit (inter-neuronal) recurrence in gradient calculations drastically reduces computational complexity to O(k^2) time and memory, making online learning feasible for larger networks.
  • The benefits of recurrent connections in processing temporal information can be maintained in Spiking Neural Networks even when their contribution to gradient computation is simplified.

Summary by alphaXiv.

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