Asynchronous Gradient Push

We consider a multi-agent framework for distributed optimization where each agent has access to local smooth strongly convex function, and the collective goal is achieve consensus on parameters that minimize sum of agents' functions. propose an algorithm wherein operates asynchronously independently other agents. When functions are strongly-convex with Lipschitz-continuous gradients, we show iterates at converge neighborhood global minimum, size depends degree asynchrony in network. agents work same rate, convergence minimizer achieved. Numerical experiments demonstrate Asynchronous Gradient-Push can objective faster than state-of-the-art synchronous first-order methods, more robust failing or stalling agents, scales better network size.