A Stratified Analysis of Bayesian Optimization Methods
This work provides a flexible framework for researchers to analyze Bayesian optimization methods, though it is incremental as it builds on existing empirical analysis approaches without introducing new algorithms.
The paper tackled the problem of empirically comparing Bayesian optimization methods by defining two metrics and a ranking mechanism for performance evaluation across different strata of test functions, which mimic hyperparameter optimization complexity but allow rapid evaluation.
Empirical analysis serves as an important complement to theoretical analysis for studying practical Bayesian optimization. Often empirical insights expose strengths and weaknesses inaccessible to theoretical analysis. We define two metrics for comparing the performance of Bayesian optimization methods and propose a ranking mechanism for summarizing performance within various genres or strata of test functions. These test functions serve to mimic the complexity of hyperparameter optimization problems, the most prominent application of Bayesian optimization, but with a closed form which allows for rapid evaluation and more predictable behavior. This offers a flexible and efficient way to investigate functions with specific properties of interest, such as oscillatory behavior or an optimum on the domain boundary.