Title:Resource-aware Exact Decentralized Optimization Using Event-triggered Broadcasting

Abstract:This work addresses the decentralized optimization problem where a group of agents with coupled private objective functions work together to exactly optimize the summation of local interests. Upon modeling the decentralized problem as an equality-constrained centralized one, we leverage the linearized augmented Lagrangian method (LALM) to design an event-triggered decentralized algorithm that only requires light local computation at generic time instants and peer-to-peer communication at sporadic triggering time instants. The proposed method is universal in the sense that it perfectly accommodates all types of convex objective functions with tailored local implementations and competitive convergence rates, that is, a rate of $O(\frac{1}{k})$ for nonsmooth objectives, the same rate for smooth objectives with a simplified iteration scheme, and a linear convergence rate for smooth and strongly convex objectives, provided that some mild conditions are satisfied. We examine the developed strategy in a decentralized logistic regression problem; comparison results illustrate its effectiveness and superiority in exploiting communication resources.