NysAct uses an eigenvalue-shifted Nyström approximation of the activation covariance to build a cheap gradient preconditioner, delivering KFAC-competitive accuracy at a fraction of the compute.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
NysAct: A Scalable Preconditioned Gradient Descent using Nystrom Approximation
NysAct uses an eigenvalue-shifted Nyström approximation of the activation covariance to build a cheap gradient preconditioner, delivering KFAC-competitive accuracy at a fraction of the compute.