The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents
arXiv: 2402.03220 · Summary page · arXiv
What changed between two versions of a paper. Changed results and numbers (before and after), conclusions, method, data, errors fixed, or only the writing, and whether the main results changed. And an email when a paper you follow gets a new version.