NeurIPS 2020

Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective Bayesian Optimization

Meta Review

*PROS: theoretical guarantees of proposed approach: calculation of unbiased gradients and SAA which is asymptotically correct, extensive evaluation. *CONS: lack of novelty: using an already known acquisition function, expected hyper-volume improvement, which is adapted for quick evaluation in a parallel setting with a Monte Carlo approximation and exact gradient computation Meta-reviewer recommendations: All the reviewers clearly agree on acceptance. I recommend the authors to take into account the reviewers' comments to improve the paper for the final version.