Motivation:
- Most ML moderns become statistics once they are trained. Or one has to retrain the model to keep it adaptive to the new environment.
Main idea:
- Instead of updating weights directly, update the rules (determined by the Hebbian coefficients) that update the weights
Some details:
- Weights are updated by Hebbian ABCD model,
- Let
be the vectorize Hebbian coefficients,
, where
and
is a fitnness evalution of
(This I am not completely certained how it is evaluated)
Ref: video, paper, twitter posts