Regret Bounds for Noise-Free Cascaded Kernelized Bandits

TL;DR

This paper develops regret bounds for optimizing noise-free, structured functions in RKHS using cascaded kernelized bandits, proposing algorithms with near-optimal theoretical guarantees.

Contribution

It introduces GPN-UCB and a non-adaptive sampling method for structured RKHS functions, providing matching upper and lower regret bounds.

Findings

GPN-UCB achieves regret bounds similar to black-box settings.

The methods have near-optimal dependence on parameters like RKHS norm and network length.

Theoretical bounds are established for both simple and cumulative regret.

Abstract

We consider optimizing a function network in the noise-free grey-box setting with RKHS function classes, where the exact intermediate results are observable. We assume that the structure of the network is known (but not the underlying functions comprising it), and we study three types of structures: (1) chain: a cascade of scalar-valued functions, (2) multi-output chain: a cascade of vector-valued functions, and (3) feed-forward network: a fully connected feed-forward network of scalar-valued functions. We propose a sequential upper confidence bound based algorithm GPN-UCB along with a general theoretical upper bound on the cumulative regret. In addition, we propose a non-adaptive sampling based method along with its theoretical upper bound on the simple regret for the Mat\'ern kernel. We also provide algorithm-independent lower bounds on the simple regret and cumulative regret. Our regret bounds for GPN-UCB have the same dependence on the time horizon as the best known in the vanilla black-box setting, as well as near-optimal dependencies on other parameters (e.g., RKHS norm and network length).