Bayesian Optimization for Estimation of Unknown Multidimensional Psychophysical Functions

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K03375
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 10040:Experimental psychology-related
• Research Institution: Osaka Electro-Communication University
• Principal Investigator: Komori Masashi 大阪電気通信大学, 情報通信工学部, 教授 (60352019)
• Co-Investigator(Kenkyū-buntansha): 遠里 由佳子 立命館大学, 情報理工学部, 教授 (80346171)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: ガウス過程選好学習 / ガウス過程回帰 / 効用関数 / 顔知覚

Outline of Final Research Achievements

In this project, we developed methodologies for estimating a multidimensional psychophysical function (referred to as a utility function) using Gaussian process regression. We employed Gaussian process preference learning (GPPL), an extension of Gaussian process regression, to estimate human utility functions based on responses to alternative two-choice tasks. Furthermore, we developed various methodologies to apply Gaussian process preference learning to various psychological problems such as facial impression researches and design researches, and demonstrated the effectiveness of the proposed method through experimental investigations. We also developed applications for conducting these experiments and explored the field of application of the methodologies.

Free Research Field

実験心理学

Academic Significance and Societal Importance of the Research Achievements

研究成果の学術的意義は（1）ガウス過程選好学習（GPPL）が多様な心理学的な問題に適用可能であり．また多次元の心的な効用関数の推定において従来の手法より高い予測精度を持つことを示したこと，（2）また効用関数の特徴や信頼性を記述するための様々な手法を確立したことである．社会的意義は，GPPLにもとづくベイズ最適化を簡便に行うことができるアプリケーション・実験システムを構築し，この手法が，言語化が容易ではない感性の可視化（他者に対する偏見の可視化，商品コンセプトの可視化）に有効であることを示したことである．