DSEBO: Automated Random Embedding for High-Dimensional Bayesian Optimization

A new method called Dynamic Shared Embedding Bayesian Optimization (DSEBO) addresses the challenge of unknown effective dimension in high-dimensional Bayesian optimization. Traditional random embedding requires prior knowledge of the effective dimension to select the subspace dimensionality, often relying on expert input or costly trial-and-error. DSEBO starts with a low-dimensional subspace and dynamically switches to a higher dimension if solutions in the current subspace are unsatisfactory. This automated approach eliminates the need for pre-specifying the effective dimension, improving efficiency and performance. The paper is available on arXiv under identifier 2605.23473.