情報処理学会 第84回全国大会 会期:2022年3月3日~5日 情報処理学会 第84回全国大会 会期:2022年3月3日~5日

2ZM-07
Study for the exploration-exploitation strategy of human based on restless two-armed bandit task
○田 家興,池田和司,吉本潤一郎,日永田智絵(奈良先端大),大平英樹(名大),木村健太(産総研)
Studying human decision-making mechanisms and modeling them can understand and predict decision-making behaviors. The core problem in our unstable environment is the exploration-exploitation dilemma, and we divide the decision-making models into two parts: (a) value function through reward; (b) strategies balance the exploitation and exploration process based on the value function. We investigate decision-making in a restless two-armed bandit task and use multiple methods for each part to fit the dataset. We use AIC and BIC to evaluate the best method and find that the exploration-exploitation tradeoff parameters can better classify the human choice patterns.