[2J1-GS-2] Machine learning: Gaussian process model

[2J1-GS-2-03] Bayesian optimization of a function affected by an external factor

〇Koji Kamitani Iwayama^1,2 (1. Shiga University, 2. JST, PRESTO)

Keywords:Bayesian optimization, Gaussian process

A sequential black-box optimization problem, where a function is affected by random external factors. In this problem, a function value may change in each round due to external effects, even if the same action is selected. The optimization of cultivation management is a prominent example of such problem. In addition to cultivation management, the weather during cultivation also affects the yield. In this example, cultivation management and weather conditions represent the action and the external factor, respectively. These problems are formulated as an optimization of the expected value under the influence of external factors, and the Gaussian process upper confidence bound of the expected value (GP-UCB-Ex) algorithm is developed. The simulation results show that the GP-UCB-Ex algorithm achieves better optimization performance compared with the ordinary GP-UCB algorithm.