Preliminary Discussion on Dynamic Causal Bayesian Optimisation with Prior Knowledge
Lists
Only you see this
Abstract
This paper proposes DCπBO, an approach to Dynamic Causal Bayesian Optimisation with Prior Knowledge, which outperforms existing state-of-the-art approaches in Dynamic Causal Bayesian Optimisation (DCBO) by enabling domain experts to specify prior knowledge. DCBO is particularly effective for systems that evolve with causal and dynamic relationships, such as complex economic models, which we want to optimise based on actual datasets. DCπBO — our proposal — extends DCBO by enabling domain experts to express a prior in the form of a function, which specifies where they believe the location of the optima to be, and it bases the search for optima on evaluations of such a given function. When the domain expert provides accurate priors, DCπBO outperforms competing approaches in our extensive experimental analysis— which includes both synthetic and real datasets — while producing comparable results when imprecise priors are provided.<br>
BibTeX
@inproceedings{lavazza2023preliminary,
author = {Luca Lavazza and Federico Cerutti},
title = {{Preliminary Discussion on Dynamic Causal Bayesian Optimisation with Prior Knowledge}},
booktitle = {Zenodo},
year = {2023},
doi = {10.5281/zenodo.8158931},
}Loading related content...