A gradient method alternating short GD steps and long Polyak steps achieves local linear convergence for overparameterized GMMs under mixture-weight assumptions.
Two models of double descent for weak features
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Local linear convergence of gradient methods for overparameterized Gaussian mixtures
A gradient method alternating short GD steps and long Polyak steps achieves local linear convergence for overparameterized GMMs under mixture-weight assumptions.