A new first-order optimizer wrapper projects gradients away from the estimated minimum-curvature direction and is claimed to stabilize transformer training, though evidence is preliminary.
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
CONDITIONAL 1representative citing papers
citing papers explorer
-
Dimer-Enhanced Optimization: A First-Order Approach to Escaping Saddle Points in Neural Network Training
A new first-order optimizer wrapper projects gradients away from the estimated minimum-curvature direction and is claimed to stabilize transformer training, though evidence is preliminary.