Optimal Regret Is Achievable with Bounded Approximate Inference Error: An Enhanced Bayesian Upper Confidence Bound Framework

This repository contains code related to the NeurIPS 2023 paper "Optimal Regret Is Achievable with Bounded Approximate Inference Error: An Enhanced Bayesian Upper Confidence Bound Framework" by Ziyi Huang, Henry Lam, Amirhossein Meisami, and Haofeng Zhang.

Paper link: https://arxiv.org/abs/2201.12955

Citation

If you find this repository or the ideas presented in our paper useful, please consider citing the following:

@inproceedings{
huang2023optimal,
  title={Optimal Regret Is Achievable with Bounded Approximate Inference Error: An Enhanced Bayesian Upper Confidence Bound Framework},
  author={Ziyi Huang and Henry Lam and Amirhossein Meisami and Haofeng Zhang},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
  year={2023},
}

Implementation

To implement the EBUCB algorithm, run EBUCB.py
To implement the algorithms in the worst-case scenarios (with only one bounded alpha-divergence), run worst_case.py