Proposal of Value Function Implementing Adaptive Cognitive Properties and Its Application to Large-Scale Computing

2015 Fiscal Year Final Research Report

• Project/Area Number: 25730150
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Soft computing
• Research Institution: Tokyo Denki University
• Principal Investigator: Takahashi Tatsuji 東京電機大学, 理工学部, 准教授 (00514514)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: 限定合理性 / n本腕バンディット問題 / 満足化 / 強化学習 / 計算論的合理性

Outline of Final Research Achievements

With the increasing amount of data and the progress in robotics, it is one of the urgent issues to establish a more efficient action selection algorithm while it learns causal relationship under uncertainty. In this study, we prove the effectiveness of a value function, the loosely symmetric (LS) model, that models the causal intuition of humans. In the multi-armed bandit problems, robotic action acquisition task, and Monte Carlo tree search, we showed the efficiency realized by the LS model. We also generalized the model loosening the original restrictions based on new analyses, and tested its cognitive validity by a meta-analysis and experiments.

Free Research Field

認知科学、知能情報学