no code implementations • 15 Mar 2024 • Yanan Bo, Yongqiang Wang
More specifically, we propose a stochastic quantization scheme and prove that it can effectively escape saddle points and ensure convergence to a second-order stationary point in distributed nonconvex optimization.