This paper introduces a 'gradient glass' model of ReLU-induced loss landscape non-smoothness and an optimizer, Alice, that combines Hessian and glass curvature, yielding modest gains over Adam on several benchmarks.
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
-
Curvature in the Looking-Glass: Optimal Methods to Exploit Curvature of Expectation in the Loss Landscape
This paper introduces a 'gradient glass' model of ReLU-induced loss landscape non-smoothness and an optimizer, Alice, that combines Hessian and glass curvature, yielding modest gains over Adam on several benchmarks.