First high-probability bounds for SignSGD with batching or majority voting under (L0, L1)-smoothness and heavy-tailed noise, with near-optimal epsilon-dependencies.
Stochastic gradient descent tricks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Sign Operator for Coping with Heavy-Tailed Noise in Non-Convex Optimization: High Probability Bounds Under $(L_0, L_1)$-Smoothness
First high-probability bounds for SignSGD with batching or majority voting under (L0, L1)-smoothness and heavy-tailed noise, with near-optimal epsilon-dependencies.