Sakana AI's PC-ALM trains 1000-layer nets without backprop; researcher notes 2018 precedent

PMinervini · x · 2026-09-15

Sakana AI introduced PC-ALM, a local-learning alternative to backpropagation that trains 1000-layer neural nets using only local dynamics, motivated by the brain's inability to implement backprop exactly. Researcher Pierce Minervini countered that the idea isn't new, citing the 2018 paper "Backpropagation and Biological Plausibility", which derived biologically plausible local algorithms via Lagrangian saddle-point search.

Original post →

More from Research

Research channel →