Gradient EM converges exponentially to optimal population loss minimizers for agnostic fitting of k parametric functions under strong convexity and smoothness of the loss, proper initialization, and separation conditions.
Gradient descent with random initialization: Fast global convergence for nonconvex phase retrieval.Mathematical Programming, 176:5–37, 2019
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Expectation Maximization (EM) Converges for General Agnostic Mixtures
Gradient EM converges exponentially to optimal population loss minimizers for agnostic fitting of k parametric functions under strong convexity and smoothness of the loss, proper initialization, and separation conditions.