Reproduce “Biologically Plausible Learning via Bidirectional Spike-Based Distillation” 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
Changze Lv, Yifei Wang, Yanxun Zhang, Yiyang Lu, Jingwen Xu, Xiaohua Wang, Di Yu, Xin Du and 2 more
A novel learning algorithm, Bidirectional Spike-based Distillation (BSD), is proposed to integrate the brain's bidirectional perception and recall processes. It satisfies five biological plausibility criteria, including spike-based communication and unsigned error signals, demonstrating performance competitive with backpropagation across image classification, sequential regression, and generative tasks using Spiking Neural Networks (SNNs).
Key insights
- Integrating bidirectional information flow (perception and recall) via mutual spike-based distillation allows for biologically plausible learning that can achieve performance comparable to backpropagation.
- Local, unsigned error signals, specifically the Relaxed Contrastive (ReCo) loss, are effective for aligning neuronal representations without requiring global error propagation or symmetric feedforward and feedback weights.
- Spiking Neural Networks (SNNs) can achieve high computational performance while adhering to strict biological plausibility criteria, demonstrating significant energy efficiency gains for inference compared to ANNs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.