Shuffling the Data, Stretching the Step-size: Sharper Bias in constant step-size SGD

Researchers have demonstrated that combining Random Reshuffling with Richardson--Romberg extrapolation improves the performance of stochastic gradient methods with constant step size, reducing bias and enhancing mean-squared error guarantees. This synergy is particularly beneficial for tasks like adversarial robustness and multi-agent learning, offering a more efficient approach to solving variational inequality problems.

Read the full article at arXiv cs.LG (ML)

---

Want to create content about this topic? Use Nemati AI tools to generate articles, social posts, and more.