Biologically Plausible Learning via Bidirectional Spike-Based Distillation

Researchers have introduced Bidirectional Spike-based Distillation (BSD), a novel algorithm that trains spiking neural networks to perform as well as those using standard error backpropagation. BSD jointly trains a feedforward network for perception and a backward network for memory recall, using only spikes for communication. Extensive benchmarks in image recognition, generation, and sequential regression show it achieves comparable performance, marking a significant step toward efficient, brain-like machine learning. Code is publicly available.