A completely uncoupled learning algorithm for general utility maximization
S. Ramakrishnan, Venkatesh Ramaiyan · 2016
In this paper, we study completely uncoupled learning algorithms for general utility maximization. We illustrate the algorithm with a wireless network application viz distributed user association. Our main contribution is expansion of achievable rate region by allowing time sharing of resources, which the previous works based on completely uncoupled strategies have ignored. First, we present a distributed user association algorithm based on a state space expansion that can achieve any desired throughput vector in the rate region of the wireless network. Then, for concave utility functions, we present a stochastic gradient algorithm with fewer synchronization requirements than known references.