Reproduce “Federated Martingale Posterior Samping” 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
Boning Zhang, Matteo Zecchin, Mingzhao Guo, Dongzhu Liu, Osvaldo Simeone
FMP enables federated image classification by sending learned, compact summaries from clients to a server, which generates hypothetical observations and refits a classifier to approximate martingale-posterior samples. On homogeneous CIFAR-10, it reaches 78.91% accuracy and 0.057 ECE, versus 77.40% and 0.142 for CFMP, with lower reported per-client communication; these comparisons cover evaluated one-shot methods, not centralized methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.