Reproduce “On Johnson-Lindenstrauss Transforms for Byzantine-Robust Optimization” 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
Aleksandr Shestakov, Vladimir Aletov, Aleksandr Beznosikov
Grad-BANT compresses distributed workers’ gradients with random projections that approximately preserve distances and inner products, allowing a server to score updates against a trusted reference direction. In a ResNet-18/CIFAR-10 experiment with seven of ten clients Byzantine, it reaches 0.827 best accuracy under IPM at 50% compression. Under ALIE it reaches 0.539 and fails to reach 60% accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.