Debiasing Conditional Stochastic Optimization. (arXiv:2304.10613v1 [cs.LG])

In this paper, we study the conditional stochastic optimization (CSO) problem
which covers a variety of applications including portfolio selection,
reinforcement learning, robust learning, causal inference, etc. The
sample-averaged gradient of the CSO objective is biased due to its nested
structure and therefore requires a high sample complexity to reach convergence.
We introduce a general stochastic extrapolation technique that effectively
reduces the bias. We show that for nonconvex smooth objectives, combining this
extrapolation with variance reduction techniques can achieve a …

algorithms applications arxiv bias causal inference complexity convergence cso etc general gradient inference optimization paper portfolio reinforcement reinforcement learning stochastic study variance