Reproduce “StratFL: Countering Covert Model Poisoning Attacks with Strategic Federated Learning” 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
Manoj Twarakavi, Venkata Sriram Siddhardh Nadendla
StratFL models federated learning as a repeated game. The server and attacker infer choices from private validation-loss changes and adjust client selection or poisoning. In ten-client, 50-round MNIST experiments, the authors report accuracy above 99% through 0.7 compromised clients; StratFL beats compared baselines through that fraction, while FLTrust does better at larger fractions. Evidence is limited to tested image-classification settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.