AI and the World's Hard Problems · Physics · arXiv · 2026-10-20
Toward a Physics of Deep Learning and Brains
The authors show that equations describing neuronal avalanches in brains also apply to cascades of activity in deep neural networks. The framework predicts that networks learn best when poised between absorbing and active phases, operating in a quasi-critical regime. Across networks trained from different initializations, maximal susceptibility proved a more reliable predictor of learning than proximity to the critical point.