Refined analysis of Clipped SGD yields improved high-probability rates involving generalized effective dimension and establishes matching lower bounds showing optimality for convergence in expectation under heavy-tailed noise.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Clipped Gradient Methods for Nonsmooth Convex Optimization under Heavy-Tailed Noise: A Refined Analysis
Refined analysis of Clipped SGD yields improved high-probability rates involving generalized effective dimension and establishes matching lower bounds showing optimality for convergence in expectation under heavy-tailed noise.