Distributed Resource Allocation for Multi-Agent Networks

@article{shorinwa_distributed_nodate,
  title = {Distributed {Resource} {Allocation} for {Multi}-{Agent} {Networks}},
  abstract = {We present a distributed algorithm for resource allocation problems where each agent computes its optimal resource allocation locally without knowing the resource allocation, objective, and constraints of other agents, guaranteeing the privacy of each agent’s local data and resource allocation. Each agent communicates with its neighbors over a point-to-point communication network to satisfy the coupling constraints on the resource allocations of all agents. Our distributed algorithm, derived from the dual formulation of the problem using the consensus alternating direction method of multipliers, applies to resource allocation problems with both convex equality and inequality coupling constraints. As such, unlike many other distributed resource allocation methods, our distributed algorithm is not limited to problems with affine coupling constraints. In addition, our algorithm does not require a feasible initialization of the resource allocations for convergence to an optimal resource allocation. We demonstrate faster empirical convergence of our distributed algorithm to the optimal resource allocation compared to other distributed resource allocation algorithms, with our algorithm converging in about two orders of magnitudes fewer communication rounds.},
  language = {en},
  author = {Shorinwa, Ola and Schwager, Mac},
  note = {Under Review}
}