Gradient Descent Converges Arbitrarily Fast for Logistic Regression via Large and Adaptive Stepsizes.

Ruiqi Zhang, Jingfeng Wu, Peter L. Bartlett

Browse the full ICML paper archive.