You Jiacheng
@YouJiacheng
Joined August 2015
2.4K Following 12.6K Followers
> each value coordinate picks one
wow that makes sense
SwiLA's choices:
① Function class: a mixture of J linear regressions; each value coordinate picks one, with a learned prior.
② Optimizer: online SGD on the likelihood-based loss; each regressor takes a delta-rule step weighted by its responsibility.
J = 1 recovers DeltaNet.
Show more
