Multi-fidelity Bayesian Optimization with Max-value Entropy Search and its parallelization

[nabenabe0928]

Multi-fidelity Bayesian Optimization with Max-value Entropy Search and its parallelization

Main points

• In principle, Fantasizing + max-value entropy search!
• Experiments used material dataset and MF version of Styblinski-Tang and Hartmann6.
• This paper defines multi-fidelity optimization (multifidelity optimization, MFO) as an optimization of an objective function such that it has various levels of fidelity in one fidelity parameter. On the other hand, single-fidelity optimization is a black-box optimization. Because normal objective functions just have only one-level in a fidelity parameter.