Expected Improvement over Stationarity-Aware Objectives for Bayesian Optimization

This repository provides the implementation of Expected Improvement over Stationarity-Aware Objectives for Bayesian Optimization from https://arxiv.org/pdf/2601.21357. It supports running the Hartmann (6d) synthetic benchmark with EI-family acquisition functions (EI, LogEI, EI-GN) under a reduced evaluation budget.

Installation

Tested with Python 3.11.9.

pip install -r requirements.txt

Running Hartmann (6d)

Run EI, LogEI, and EI-GN via:

python main.py --config ./configs/hartmann_EI.yaml
python main.py --config ./configs/hartmann_LogEI.yaml
python main.py --config ./configs/hartmann_EIGN.yaml

Cite This Work

If you find this work useful, please consider citing:

@article{ip2026expected,
  title={Expected Improvement via Gradient Norms},
  author={Ip, Joshua Hang Sai and Makrygiorgos, Georgios and Mesbah, Ali},
  journal={arXiv preprint arXiv:2601.21357},
  year={2026}
}