Towards Self-Adjusting Weighted Expected Improvement for Bayesian Optimization

, , , , and .
GECCO '23: Proceedings of the Genetic and Evolutionary Computation Conference Companion, Association for Computing Machinery Special Interest Group on Genetic and Evolutionary Computation (SIGEVO), (2023)


In optimization, we often encounter expensive black-box problemswith unknown problem structures. Bayesian Optimization (BO) isa popular, surrogate-assisted and thus sample-efficient approachfor this setting. The BO pipeline itself is highly configurable withmany different design choices regarding the initial design, surrogatemodel and acquisition function (AF). Unfortunately, our understand-ing of how to select suitable components for a problem at hand isvery limited. In this work, we focus on the choice of the AF, whosemain purpose it is to balance the trade-off between exploring re-gions with high uncertainty and those with high promise for goodsolutions. We propose Self-Adjusting Weighted Expected Improve-ment (SAWEI), where we let the exploration-exploitation trade-offself-adjust in a data-driven manner based on a convergence crite-rion for BO. On the BBOB functions of the COCO benchmark, ourmethod performs favorably compared to handcrafted baselines andserves as a robust default choice for any problem structure. WithSAWEI, we are a step closer to on-the-fly, data-driven and robustBO designs that automatically adjust their sampling behavior tothe problem at hand.



  • @ail3s
  • @dblp

Comments and Reviews