8
28

Thompson Sampling with Approximate Inference

Abstract

We study the effects of approximate inference on the performance of Thompson sampling in the kk-armed bandit problems. Thompson sampling is a successful algorithm for online decision-making but requires posterior inference, which often must be approximated in practice. We show that even small constant inference error (in α\alpha-divergence) can lead to poor performance (linear regret) due to under-exploration (for α<1\alpha<1) or over-exploration (for α>0\alpha>0) by the approximation. While for α>0\alpha > 0 this is unavoidable, for α0\alpha \leq 0 the regret can be improved by adding a small amount of forced exploration even when the inference error is a large constant.

View on arXiv
Comments on this paper