Skip to yearly menu bar Skip to main content


Poster

Thompson Sampling for High-Dimensional Sparse Linear Contextual Bandits

Sunrit Chakraborty · Saptarshi Roy · Ambuj Tewari

Exhibit Hall 1 #630
[ ]
[ PDF [ Poster

Abstract:

We consider the stochastic linear contextual bandit problem with high-dimensional features. We analyze the Thompson sampling algorithm using special classes of sparsity-inducing priors (e.g., spike-and-slab) to model the unknown parameter and provide a nearly optimal upper bound on the expected cumulative regret. To the best of our knowledge, this is the first work that provides theoretical guarantees of Thompson sampling in high-dimensional and sparse contextual bandits. For faster computation, we use variational inference instead of Markov Chain Monte Carlo (MCMC) to approximate the posterior distribution. Extensive simulations demonstrate the improved performance of our proposed algorithm over existing ones.

Chat is not available.