Fast and Efficient Asynchronous Gossip Algorithm for Robust and Non-Smooth Convex Decentralized Learning
AsylADMM is a novel asynchronous gossip algorithm designed for decentralized non-smooth convex optimization on resource-constrained devices. It achieves efficiency by requiring only two variables per node, simplifying updates, and eliminating neighbor storage. The method converges faster than existing baselines on median and quantile estimation problems while proving robustness against outliers.