Source code for the AISTATS 2024 paper
"Looping in the Human: Collaborative and Explainable Bayesian Optimization" paper
We prepared an example of CoExBO with battery example.
- Demo1 human feedback for battery experiments.ipynb
- Demo2 synthetic human response.ipynb
Fig. 1: In Collaborative and Explainable Bayesian Optimization (CoExBO), a human expert collaborates with BO to refine electrolyte materials. While experts excel in discerning material differences rather than identifying the best one, pairwise comparisons and explanations boost their feedback accuracy and trust. This guides the BO to produce better candidates, en- suring quicker convergence.
Collaborative and Explainable BO (CoExBO)
- BO combines experimental results and expert preferences.
- BO generates pairwise candidates along with explanations.
- Human interprets the acquisitions and picks their preferred candidate
- Human conducts experiments and repeat step 1.
Fig. 2: Figure 2: Explanation flow: Spatial relation: BO visualizes the surrogate model’s predictive distribution and estimated human preference models for the two primary dimensions determined by Shapley values. Feature importance: Users’ values are provided for both candidates’ predictive mean, standard deviation, and acquisition function. Selection accuracy feed- back: After observing the function value, a post-hoc evaluation of the correct selection probability is given.
Utilising GP-SHAP, we can provide insights into the undergoing of the BO by attributing feature importance to the followings:
- Surrogate GP model
- Acquisition function (GP-UCB)
botorch 0.8.4 gpytorch 1.10 torch 1.13.0
Please cite this work as
@inproceedings{adachi2024looping,
title={Looping in the Human: Collaborative and Explainable Bayesian Optimization},
author={Adachi, Masaki and Planden, Brady and Howey, David and Osborne, Michael A and Orbell, Sebastian and Ares, Natalia and Muandet, Krikamol and Chau, Siu Lun},
booktitle={International Conference on Artificial Intelligence and Statistics},
pages={505--513},
year={2024},
organization={PMLR}
}