Reproduce “Warming Up for Zeroth-Order Federated Pre-Training with Low Resource Clients” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Gwen Legate, Irina Rish, Eugene Belilovsky

ZOWarmUp trains shared federated models with clients whose resource limits prevent conventional updates. It first uses high-resource clients for conventional training, then switches to zeroth-order updates based on paired forward-pass losses and seed/scalar exchanges. In simulated CIFAR-10 with ResNet18 and a 10/90 client split, it reports 54.3% accuracy versus 44.3% when low-resource clients are excluded, a result from one tested setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.