Under the Polyak-Lojasiewicz condition, a proximal primal-dual algorithm and an AdaGrad-style variant maximize AUC with deep networks at O~(1/epsilon) sample complexity, with adaptive iteration complexity under slow cumulative gradient growth.
Universal stagewise learning for non-convex problems with convergence on averaged solutions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Stochastic AUC Maximization with Deep Neural Networks
Under the Polyak-Lojasiewicz condition, a proximal primal-dual algorithm and an AdaGrad-style variant maximize AUC with deep networks at O~(1/epsilon) sample complexity, with adaptive iteration complexity under slow cumulative gradient growth.