Improved Incremental Constraint Projection Method by Sampling Multiple Constraints
Jialin Liu, Yuantao Gu, Mengdi Wang · 2014
We focus on stochastic optimization problem and variational inequalities problem in which the objective function is expectation of a randomized convex real-value func-tion, and the feasible set is the intersection of large number convex sets. We propose ICPM-SMC as a general framework which involves Incremental Constraints Projection Method (ICPM) as one of its special schemes. Then we analyze its convergence rate and convergence stability. For rate, we uses different expected errors as criteria for different cases; for stability, we propose three criteria. Use these tools, we find that we can con-trol the parameters and sampling scheme in ICPM-SMC to get particular convergence