JSAI2019

Presentation information

General Session

General Session » [GS] J-2 Machine learning

[4I3-J-2] Machine learning: analysis and buliding of basic models

Fri. Jun 7, 2019 2:00 PM - 3:20 PM Room I (306+307 Small meeting rooms)

Chair:Takuya Hiraoka Reviewer:Shohei Higashiyama

2:00 PM - 2:20 PM

[4I3-J-2-01] On the adaptive selection of the hyperparameter of aquisition function based on the sample point gradient.

〇Tatsuya Hasebe1, Ichiro Kataoka1 (1. Hitachi, Ltd.)

Keywords:Bayesian Optimization , Gaussian Process, Bandits, Acquisition function

In order to easily determine the optimal parameter for acquisition function of bayesian optimization,
the adaptive selection criterion for GP-UCB acquisition function parameter based on the sample point gradient is proposed.
As a result, an equivalent or superior performance of the proposed method is confirmed by comparing the cummulative regret.
Moreover, the round number dependence of the adaptively determined prameter is consistent with the trend of the parameter of GP-UCB which is theoretically derived.