Distributed Stochastic Dual Subgradient for Constraint-Coupled Optimization

In this letter we consider a distributed stochastic optimization framework in which agents network aim to cooperatively learn an optimal network-wide policy. The goal is compute local functions minimize the expected value of given cost, subject individual constraints and average coupling constraints. order handle challenges context, resort Lagrangian duality approach that allows us derive associated dual problem with separable structure. Thus, propose algorithm, without central coordinator, exploits consensus iterations approximation find solution problem, attractive scalability properties. We demonstrate convergence proposed scheme validate its behavior through simulations.