Adaptive Gradient Descent without Descent
arXiv:1910.09529
Abstract
We present a strikingly simple proof that two rules are sufficient to automate gradient descent: 1) don't increase the stepsize too fast and 2) don't overstep the local curvature. No need for functional values, no line search, no information about the function except for the gradients. By following these rules, you get a method adaptive to the local geometry, with convergence guarantees depending only on the smoothness in a neighborhood of a solution. Given that the problem is convex, our method converges even if the global smoothness constant is infinity. As an illustration, it can minimize arbitrary continuously twice-differentiable convex function. We examine its performance on a range of convex and nonconvex problems, including logistic regression and matrix factorization.
References in corpus (9)
- ADADELTA: An Adaptive Learning Rate Method
- On the Convergence of Adam and Beyond
- AdaGrad stepsizes: Sharp convergence over nonconvex landscapes
- Optimized first-order methods for smooth convex minimization
- Error Feedback Fixes SignSGD and other Gradient Compression Schemes
- Adaptive Bound Optimization for Online Convex Optimization
- Unified Optimal Analysis of the (Stochastic) Gradient Method
- Adaptive restart of accelerated gradient methods under local quadratic growth condition
- Revisiting the Polyak step size