1 December 2023 in Transactions of Nonferrous Metals Society of China
12 April 2021 in IEEE Transactions on Parallel and Distributed Systems
Decentralized dual methods play significant roles in large-scale optimization, which effectively resolve many constrained optimization problems in machine learning and power systems. In this article, we focus on studying a class of totally non-smooth constrained composite optimization problems over multi-agent systems, where the mutual goal of agents in the system is to optimize a sum of two separable non-smooth functions consisting of a strongly-convex function and another convex (not necessarily strongly-convex) function. Agents in the system conduct parallel local computation and communication in the overall process without leaking their private information. In order to resolve the totally non-smooth constrained composite optimization problem in a fully decentralized manner, we devise a synchronous decentralized dual proximal (SynDe-DuPro) gradient algorithm and its asynchronous version (AsynDe-DuPro) based on the randomized block-coordinate method. Both SynDe-DuPro and AsynDe-DuPro algorithms are theoretically proved to achieve the globally optimal solution to the totally non-smooth constrained composite optimization problem relied on the quasi-Fejér monotone theorem. As a main result, AsynDe-DuPro algorithm attains the globally optimal solution without requiring all agents to be activated at each iteration and thus is more robust than most existing synchronous algorithms. The practicability of the proposed algorithms and correctness of the theoretical findings are demonstrated by the experiments on a constrained Decentralized Sparse Logistic Regression (DSLR) problem in machine learning and a Decentralized Energy Resources Coordination (DERC) problem in power systems.
28 November 2022 in IEEE Transactions on Network Science and Engineering
Distributed stochastic optimization, arising in the crossing and integration of traditional stochastic optimization, distributed computing and storage, and network science, has advantages of high efficiency and a low per-iteration computational complexity in resolving large-scale optimization problems. This paper concentrates on resolving a large-scale convex finite-sum optimization problem in a multi-agent system over unbalanced directed networks. To tackle this problem in an efficient way, a distributed consensus optimization algorithm, adopting the push-sum technique and a distributed loopless stochastic variance-reduced gradient (LSVRG) method with uncoordinated triggered probabilities, is developed and named Push-LSVRG-UP. Each agent under this algorithmic framework performs only local computation and communicates only with its neighbors without leaking their private information. The convergence analysis of Push-LSVRG-UP is relied on analyzing the contraction relationships between four error terms associated with the multi-agent system. Theoretical results provide an explicit feasible range of the constant step-size, a linear convergence rate, and an iteration complexity of Push-LSVRG-UP when achieving the globally optimal solution. It is shown that Push-LSVRG-UP achieves the superior characteristics of accelerated linear convergence, fewer storage costs, and a lower per-iteration computational complexity than most existing works. Meanwhile, the introduction of an uncoordinated probabilistic triggered mechanism allows Push-LSVRG-UP to facilitate the independence and flexibility of agents in computing local batch gradients. In simulations, the practicability and improved performance of Push-LSVRG-UP are manifested via resolving two distributed learning problems based on real-world datasets.
9 February 2024 in IEEE Transactions on Signal and Information Processing over Networks
This paper investigates the generalized Nash equilibria (GNE) seeking problem for noncooperative games, where all players dedicate to selfishly minimizing their own cost functions subject to local constraints and coupled constraints. To tackle the considered problem, we initially form an explicit local equilibrium condition for its variational formulation. By employing proximal splitting operators, a novel distributed primal-dual splitting algorithm with full-decision information (Dist_PDS_FuDeIn) is designed, eliminating the need for global step-sizes. Furthermore, to address scenarios where players lack access to all other players' decisions, a local estimation is introduced to approximate the decision information of other players, and a fully distributed primal-dual splitting algorithm with partial-decision information (Dist_PDS_PaDeIn) is then proposed. Both algorithms enable the derivation of new distributed forward-backward-like extensions. Theoretically, a new analytical approach for convergence is presented, demonstrating that the proposed algorithms converge to the variational GNE of games, and their convergence rates are also proven, provided that uncoordinated step-sizes are positive and less than explicit upper bounds. Moreover, the approach not only generalizes the forward-backward splitting technique but also improves convergence rates of several well-known algorithms. Finally, the advantages of Dist_PDS_FuDeIn and Dist_PDS_PaDeIn are illustrated through comparative simulations.
24 May 2023 in Information Sciences
6 August 2024 in IEEE Transactions on Automatic Control
This paper considers a class of noncooperative games in which the feasible decision sets of all players are coupled together by a coupled inequality constraint. Adopting the variational inequality formulation of the game, we first introduce a new local edge-based equilibrium condition with full information. Considering challenges when communication delays occur, we then devise an asynchronous distributed algorithm to seek a generalized Nash equilibrium. This asynchronous scheme arbitrarily activates one player to start new computations independently at different iteration instants, which means that the picked player can use the involved out-dated information from itself and its neighbors to perform new updates. In theoretical aspect, we provide explicit conditions on algorithm parameters, for instance, the step-sizes to establish a sublinear convergence rate for the synchronous version. Next, the asynchronous algorithm guarantees almost sure convergence in expectation under the same step-size conditions and some standard assumptions. Finally, the viability and performance of the proposed algorithm are demonstrated by numerical studies on the Cournot competition.