A Hessian-based criterion, KH05, increases when a model generalizes worse to a new dataset, offering a cheap generalization estimate, but the evidence is limited and the criterion's exponent was chosen post hoc.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Investigating generalization capabilities of neural networks by means of loss landscapes and Hessian analysis
A Hessian-based criterion, KH05, increases when a model generalizes worse to a new dataset, offering a cheap generalization estimate, but the evidence is limited and the criterion's exponent was chosen post hoc.