Polyak Step Size

The Polyak step size is an adaptive learning rate method for optimization algorithms like stochastic gradient descent (SGD), aiming to automatically adjust the step size during training, eliminating the need for manual tuning. Current research focuses on extending its application to reinforcement learning and improving its robustness and convergence properties, particularly in non-convex settings and for over-parameterized models, often incorporating techniques like preconditioning and variance reduction. This research is significant because efficient and robust optimization is crucial for many machine learning applications, and the Polyak step size offers a promising approach to improve the performance and stability of various algorithms.

Papers

April 11, 2024

Enhancing Policy Gradient with the Polyak Step-Size Adaption
Yunxiang Li, Rui Yuan, Chen Fan, Mark Schmidt, Samuel Horváth, Robert M. Gower, Martin Takáč
Reinforcement Learning Policy Gradient Reinforcement Learning Algorithm Faster Convergence Polyak Step Size

October 3, 2023

Stochastic Gradient Descent with Preconditioned Polyak Step-size
Farshed Abdukhakimov, Chulu Xiang, Dmitry Kamzolov, Martin Takáč
Stochastic Gradient Descent Learning Rate Multiple Optimization Polyak Step Size

August 11, 2023

Adaptive SGD with Polyak stepsize and Line-search: Robust Convergence and Variance Reduction
Xiaowen Jiang, Sebastian U. Stich
Variance Reduction Step Size Parameterized Model Line Search Stochastic Line Search Gradient Evaluation Stochastic Polyak Polyak Step Size

July 26, 2023

Function Value Learning: Adaptive Learning Rates Based on the Polyak Stepsize and Function Splitting in ERM
Guillaume Garrigos, Robert M. Gower, Fabian Schaipp
Gradient Descent Stochastic Gradient Descent Learning Rate Empirical Risk Minimization Value Function Step Size Splitting Algorithm Polyak Step Size

February 24, 2022

Cutting Some Slack for SGD with Adaptive Polyak Stepsizes
Robert M. Gower, Mathieu Blondel, Nidham Gazagnadou, Fabian Pedregosa
Loss Function Stochastic Gradient Descent Stochastic Gradient Step Size Adaptive Method Polyak Step Size

Polyak Step Size

Papers

Enhancing Policy Gradient with the Polyak Step-Size Adaption

Stochastic Gradient Descent with Preconditioned Polyak Step-size

Adaptive SGD with Polyak stepsize and Line-search: Robust Convergence and Variance Reduction

Function Value Learning: Adaptive Learning Rates Based on the Polyak Stepsize and Function Splitting in ERM

Cutting Some Slack for SGD with Adaptive Polyak Stepsizes