The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents

Open Webpage

Yatin Dandi, Emanuele Troiani, Luca Arnaboldi, Luca Pesce, Lenka Zdeborová, Florent Krzakala

Published: 01 Jan 2024, Last Modified: 15 May 2025ICML 2024EveryoneRevisionsBibTeXCC BY-SA 4.0