Hebbian Weight Updates

The precision-learning experiment indicates that better error conditioning does not automatically yield more stable weight updates at depth. The gHGF formulation produces Hebbian-compatible weight-update rules that use precision as a natur…

1 sources - 4 claims

The precision-learning experiment indicates that better error conditioning does not automatically yield more stable weight updates at depth. The gHGF formulation produces Hebbian-compatible weight-update rules that use precision as a natural-gradient weight. No current Hebbian rule is fully consistent with volatility-coupled precision dynamics. Deriving weight-update rules consistent with volatility-coupled precision dynamics is the principal open research problem.