Modelling of human decision making using inverse optimization

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K04455
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 21040:Control and system engineering-related
• Research Institution: Kyushu University
• Principal Investigator: Murata Junichi 九州大学, システム情報科学研究院, 教授 (60190914)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 人のモデル / 人の判断 / 社会サービス / 個人化 / 逆強化学習 / 多目的最適化 / 対話型進化計算

Outline of Final Research Achievements

Humans decide their actions and select the best one among available alternatives based on their own judgment criteria. The criteria can be utilized, for example, to design machines that compete with skilled craftsmen. We, however, cannot directly observe them. We only can observe the actions taken and the alternatives selected. The research proposes methods that estimate judgment criteria using those observable data. The biggest issue here is that the data only cannot determine the criteria uniquely. To solve this, the research proposed a method that determines a representation of judgment criterion with a complexity suitable for the amount of given data, and another method was developed that utilizes fluctuations in human actions, i.e., the fact that humans do not always take the best actions, as additional and useful information. Applications of these methods to example problems verified their validity.

Free Research Field

システム工学・制御工学

Academic Significance and Societal Importance of the Research Achievements

人が持つ判断基準を推定することができると，その人が好む意匠の発見支援，熟練者と同等の機能を持つ装置，人に不快感を感じさせない行動誘導，運転者個人の嗜好にあった車の自動運転などに活用することができ，大きな社会的意義を持つ．本研究で取り扱っている観測可能なデータから判断基準を推定する過程は，逆最適化問題として捉えることができる．逆最適化問題は唯一解が存在しない不良設定問題であるが，本研究では，表現方法の複雑さとデータの量のバランスをとる方法や，最適解以外の解も活用して利用する情報を増やす方法により，この問題点の解決を図っている．ここに大きな学術的意義がある．