Accelerating Approximate Thompson Sampling with Underdamped Langevin Monte Carlo

Code

• github.com/haoyangzheng1996/ts_ulmc
  OfficialIn papernone★ 4

Abstract

Approximate Thompson sampling with Langevin Monte Carlo broadens its reach from Gaussian posterior sampling to encompass more general smooth posteriors. However, it still encounters scalability issues in high-dimensional problems when demanding high accuracy. To address this, we propose an approximate Thompson sampling strategy, utilizing underdamped Langevin Monte Carlo, where the latter is the go-to workhorse for simulations of high-dimensional posteriors. Based on the standard smoothness and log-concavity conditions, we study the accelerated posterior concentration and sampling using a specific potential function. This design improves the sample complexity for realizing logarithmic regrets from O(d) to O(d). The scalability and robustness of our algorithm are also empirically validated through synthetic experiments in high-dimensional bandit problems.

Tasks

Reproductions