A new regularizer increases the condition number of the linear-probing Hessian on harmful tasks, making gradient-descent fine-tuning slower, but the theoretical analysis contains a false claim.
(4) [Monotonic Decrease] If σmax S is unique, update S with ∇SRwell(S) such that S′ =S−η 1∇SRwell(S) for 0< η1 < κ(S)−1 (1− 1 p )κ(S)+ 1 p , thenκ(S ′)< κ(S)
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
-
Model Immunization from a Condition Number Perspective
A new regularizer increases the condition number of the linear-probing Hessian on harmful tasks, making gradient-descent fine-tuning slower, but the theoretical analysis contains a false claim.