Get Started
Research questionHow do anytime constraints limit convergence rates for predetermined, possibly negative-step gradient descent in smooth convex optimization?Predetermined stepsizes are fixed in advance, but anytime schedules must provide guarantees across iteration horizons rather than only at a selected final horizon. The central difficulty is identifying the strongest convergence rates possible under these scheduling constraints.
Machine Learning
Research Paper
Latest papersRecent research connected to this question, newest first.Improved Gradient Descent Lower Bounds Beyond NesterovApplies to gradient descent for smooth convex optimization with predetermined stepsizes. The supplied evidence concerns lower bounds for both non-anytime and anytime schedules, including schedules that may use negative stepsizes.research paper · Sep 3, 2026
Related questions
For non-separable logistic regression, when does constant-step gradient descent converge globally rather than enter a stable cycle?How can Adam’s error be bounded for strongly convex stochastic optimization without assuming bounded iterates?How can iterated one-step approximations control error in strongly continuous convex monotone semigroups?What convergence guarantees and irreducible errors arise in random-reshuffling optimization for finite sums and composite objectives?
Home
Topics
Search
Library