Distributed Randomized Gradient-Free Mirror Descent Algorithm for Constrained Optimization

Distributed Constrained Optimization with Delayed Subgradient Information over Time-Varying Network under Adaptive Quantization

“…Finally, the algorithm can asymptotically converge to an optimal solution without quantization error. In this paper, we significantly improve our previous works [12,14,13] on distributed mirror descent methods in several aspects. To the best of our knowledge, this is the first work to propose the adaptive quantization method to address limited communication channel in mirror descent algorithm and simultaneously take delayed subgradient information into consideration in the study of distributed mirror descent method.…”

“…Recently, the distributed optimization algorithms for the network system have been studied widely (see in [1]- [14]). In the distributed optimization problem, there is no central coordination between different agents.…”

“…Mirror descent algorithms have been studied extensively in recent year. Here are some excellent mirror descent algorithm references (see [3,6,12,13,14]). In [3], the authors propose stochastic subgradient mirror descent algorithm for constrained distributed optimization problems.…”

“…In [13], the authors propose distributed stochastic mirror descent method over a class of time-varying network for strongly convex objective functions. In [14], the authors propose distributed randomized gradient-free mirror descent algorithm for constrained optimization.…”

“…In the distributed optimization (see [1,2,3,5,13,14,24,25,30,31,32]), each agent needs to update parameter and calculate (sub)gradient based on local parameter in parallel. Then each agent receives current (sub)gradient information.…”