Convergence analysis of gradient descent stochastic algorithms (Q1359455): Difference between revisions

The paper provides convergence analysis results for a sample-path based stochastic gradient-descent algorithm for optimizing expected-value performance measures in discrete event systems. The algorithm requires that the distance between two consecutive iteration points converge to zero as the iterate count goes to infinity. The paper gives two convergence results: one for the case where the expected-value function is continuously differentiable and the other when the function is nondifferentiable but the sample performance functions are convex. The proofs are based on a version of the uniform law of large numbers which is provable for many discrete event systems where infinitesimal perturbation analysis is known to be strongly consistent.