75
57

High-Dimensional Experimental Design and Kernel Bandits

Abstract

In recent years methods from optimal linear experimental design have been leveraged to obtain state of the art results for linear bandits. A design returned from an objective such as GG-optimal design is actually a probability distribution over a pool of potential measurement vectors. Consequently, one nuisance of the approach is the task of converting this continuous probability distribution into a discrete assignment of NN measurements. While sophisticated rounding techniques have been proposed, in dd dimensions they require NN to be at least dd, dlog(log(d))d \log(\log(d)), or d2d^2 based on the sub-optimality of the solution. In this paper we are interested in settings where NN may be much less than dd, such as in experimental design in an RKHS where dd may be effectively infinite. In this work, we propose a rounding procedure that frees NN of any dependence on the dimension dd, while achieving nearly the same performance guarantees of existing rounding procedures. We evaluate the procedure against a baseline that projects the problem to a lower dimensional space and performs rounding which requires NN to just be at least a notion of the effective dimension. We also leverage our new approach in a new algorithm for kernelized bandits to obtain state of the art results for regret minimization and pure exploration. An advantage of our approach over existing UCB-like approaches is that our kernel bandit algorithms are also robust to model misspecification.

View on arXiv
Comments on this paper