Communication-efficient Preference-based Federated Multi-resource Allocation

In many applications in communication networks, smart energy systems, edge computing, smart cities, etcetera, agents wish not to exchange information with other agents in the network and make decisions based on their choices and preferences and local computation. The agents cooperate with a central server to minimize the overall cost to the network. Such a setting is called a federated setting; it has recently received much interest from the research community. In this paper, we develop a communication-efficient stochastic multi-resource allocation algorithm for federating settings, generalizing the additive increase multiplicative decrease (AIMD) algorithm. Our solution does not require inter-agent communication. We consider a central server that coordinates with the agents to track the aggregate consumption of resources and sends one-bit feedback signals in the network when the resource capacity constraints are violated. We show the convergence of the average allocations to the optimal values through numerical results.