Finding Local Minima Efficiently in Decentralized Optimization

Neural Information Processing Systems 

In this paper we study the second-order optimality of decentralized stochastic algorithm that escapes saddle point efficiently for nonconvex optimization problems.