Distributed Stochastic Nested Optimization via Cubic Regularization

Abstract: This paper considers a nested stochastic distributed optimization problem. In it, approximate solutions to realizations of the inner-problem are leveraged to obtain a Distributed Stochastic Cubic Regularized Newton (DiSCRN) update to the decision variable of the outer problem. We provide an example involving electric vehicle users with various preferences which demonstrates that this model is appropriate and sufficiently complex for a variety of data-driven multi-agent settings, in contrast to non-nested models. The main two contributions of the paper are: (i) development of local stopping criterion for solving the inner optimization problem which guarantees sufficient accuracy for the outer-problem update, and (ii) development of the novel DiSCRN algorithm for solving the outer-problem and a theoretical justification of its efficacy. Simulations demonstrate that this approach is more stable and converges faster than standard gradient and Newton outer-problem updates in a highly nonconvex scenario, and we also demonstrate that the method extends to an EV charging scenario in which resistive battery losses and a time-of-use pricing model are considered over a time horizon.