A solution to the learning dilemma for recurrent networks of spiking neurons

In short

The authors propose that combining two specific findings from neuroscience allows spiking neuron networks to learn online using gradient descent. They show this new method, called e-prop, matches the performance of standard backpropagation while offering a path for efficient on-chip learning in spike-based hardware. This work addresses the long-standing challenge of how biological-like recurrent networks can adapt through synaptic plasticity without relying on traditional machine learning techniques.

The summaries on this page were written by a small AI model running on my own computer, from the paper's abstract. They can get details wrong; the paper is the source.

Next in the stack: EEG microstates of dreams →