Change comment

This commit is contained in:
Daniel Povey 2022-07-05 17:11:45 -07:00
parent 2692d5f903
commit 41368f6b63

View File

@ -26,9 +26,8 @@ import logging
class LearnedGradient(Optimizer):
"""
Implements 'Learned Gradient' update. It's a learned positive definite learning-rate
matrix for each dimension of each tensor, that we multiply the derivative by, and
then scale the delta/param-change before the update.
Implements 'Learned Gradient' update. It's special factorization of the
parameter matrix..
Args: