Decentralized Proximal Gradient Method for Non-convex Composite problems with Inexact Gradient
Keywords: Decentralized Optimization, Non-smooth optimization, proximal PL-condition
Abstract: Optimization problems with composite structure appears in different areas: machine learning, control, signal processing and so on. Gradient-type methods are common approach for such problems. Nevertheless, the exact gradient is not available in many pracitcal applications. Especially, it holds for decentrazed case. Therefore, we consider decentralized proximal gradient method with inexact gradient for time-variing graphs. This work contains analysis for problems with functions that satisfy proximal Polyak-Łojasiewicz condition. Thus, there is complexity estimations in terms both of oracle calls, and of communications number. Additionally, we consider stochastic case too. Besides, we provide numerical experiments to demonstrate performance of considered approach.
Submission Number: 40
Loading