An adaptive approach to Bayesian Optimization with switching costs

Stefan Pricopie,Richard Allmendinger,Manuel Lopez-Ibanez,Clyde Fare,Matt Benatan,Joshua Knowles

2024-05-15

Abstract:We investigate modifications to Bayesian Optimization for a resource-constrained setting of sequential experimental design where changes to certain design variables of the search space incur a switching cost. This models the scenario where there is a trade-off between evaluating more while maintaining the same setup, or switching and restricting the number of possible evaluations due to the incurred cost. We adapt two process-constrained batch algorithms to this sequential problem formulation, and propose two new methods: one cost-aware and one cost-ignorant. We validate and compare the algorithms using a set of 7 scalable test functions in different dimensionalities and switching-cost settings for 30 total configurations. Our proposed cost-aware hyperparameter-free algorithm yields comparable results to tuned process-constrained algorithms in all settings we considered, suggesting some degree of robustness to varying landscape features and cost trade-offs. This method starts to outperform the other algorithms with increasing switching-cost. Our work broadens out from other recent Bayesian Optimization studies in resource-constrained settings that consider a batch setting only. While the contributions of this work are relevant to the general class of resource-constrained problems, they are particularly relevant to problems where adaptability to varying resource availability is of high importance

Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to make an optimal choice between evaluation cost and setup cost in resource - constrained sequential experimental design. Specifically, when changes in certain design variables lead to switching costs, how to effectively balance the relationship between frequent evaluation and maintaining the same setup to achieve a more efficient optimization process. This is very important in practical applications. For example, in the automotive or electronics manufacturing industries, reconfiguring the production line incurs additional time and cost, so frequent adjustments are not advisable. The paper proposes an adaptive Bayesian optimization method, aiming to deal with this situation where there are switching costs. By introducing two new algorithms (one considering cost and the other ignoring cost) and improving existing algorithms, it improves the optimization performance under different switching costs. The main contribution of the paper lies in extending previous research, transforming it from a batch - constrained optimization problem to a sequential optimization problem, and proposing a cost - aware hyperparameter - free algorithm that can maintain a certain robustness under different cost trade - offs.