Communication-efficient Allocation of Multiple Indivisible Resources in a Federated Multi-agent System

A federated multi-agent system is a multi-agent system wherein agents collaborate with a central server to optimize system goals without sharing their private information. We develop a communication-efficient solution to resource allocation problems for a population of agents coupled through multiple indivisible shared resources in a federated multi-agent system. The agents demand resources in a probabilistic way based on their local computation and preferences, and the agents receive either one unit of a resource or do not receive it. The agents are not required to share their cost functions or derivatives of cost functions with other agents or the central server. Optimal control of a population of such agents, subject to capacity constraints, is widely found in many application domains, such as smart energy systems, intelligent transportation systems, and edge computing, to name a few. We present convergence results using multi-time scale stochastic approximation techniques and an example of electric vehicle charging point allocation illustrating the efficacy of the developed solution.