PolyBO: Accelerating Bayesian Optimization with Pseudo-Experimental Data

A new method called PolyBO aims to accelerate Bayesian optimization (BO) by generating high-quality pseudo-experimental data even when only limited experimental data are available. BO is a sequential optimization technique that balances exploration and exploitation, commonly used for hyperparameter tuning in deep learning. However, in practical experimental science, each evaluation can be costly and time-consuming, leading to poor convergence. Recent approaches have used pseudo-experimental data to speed up optimization, but the quality of such data suffers when few real experiments exist. PolyBO addresses this by using adaptive polynomial regression to generate reliable pseudo-data, enabling more efficient BO under a limited evaluation budget. The method is detailed in a preprint on arXiv (2607.22238).