MAC approximates KFAC's curvature with a rank-1 mean-activation matrix, making second-order training nearly as cheap as SGD while preserving accuracy.
Gradient descent on neurons and its link to approximate second-order optimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
method 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
method 1polarities
support 1representative citing papers
citing papers explorer
-
MAC: An Efficient Gradient Preconditioning using Mean Activation Approximated Curvature
MAC approximates KFAC's curvature with a rank-1 mean-activation matrix, making second-order training nearly as cheap as SGD while preserving accuracy.