2017/07/18 by Amrit Singh Bedi, Bedi, Amrit Singh, Alec Koppel +3
Computer Science · #Cooperative Communication and Network Coding #Distributed Control Multi-Agent Systems #Distributed Sensor Networks and Detection Algorithms #FOS: Mathematics #Optimization and Control (math.OC)
paper · pdf · doi:10.48550/arxiv.1707.05816
openalex publication_date 2017/07/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We consider expected risk minimization in multi-agent systems comprised of\ndistinct subsets of agents operating without a common time-scale. Each\nindividual in the network is charged with minimizing the global objective\nfunction, which is an average of sum of the statistical average loss function\nof each agent in the network. Since agents are not assumed to observe data from\nidentical distributions, the hypothesis that all agents seek a common action is\nviolated, and thus the hypothesis upon which consensus constraints are\nformulated is violated. Thus, we consider nonlinear network proximity\nconstraints which incentivize nearby nodes to make decisions which are close to\none another but not necessarily coincide. Moreover, agents are not assumed to\nreceive their sequentially arriving observations on a common time index, and\nthus seek to learn in an asynchronous manner. An asynchronous stochastic\nvariant of the Arrow-Hurwicz saddle point method is proposed to solve this\nproblem which operates by alternating primal stochastic descent steps and\nLagrange multiplier updates which penalize the discrepancies between agents.\nThis tool leads to an implementation that allows for each agent to operate\nasynchronously with local information only and message passing with neighbors.\nOur main result establishes that the proposed method yields convergence in\nexpectation both in terms of the primal sub-optimality and constraint violation\nto radii of sizes \O(\√(T)) and \O(T3/4),\nrespectively. Empirical evaluation on an asynchronously operating wireless\nnetwork that manages user channel interference through an adaptive\ncommunications pricing mechanism demonstrates that our theoretical results\ntranslates well to practice.\n