Effect of sensory prediction error on reward-based action selection

2019 Fiscal Year Final Research Report

• Project/Area Number: 18K17893
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 59030:Physical education, and physical and health education-related
• Research Institution: Tokyo University of Agriculture and Technology
• Principal Investigator: Hayashi Takuji 東京農工大学, 工学(系)研究科(研究院), 特別研究員 (80815358)
• Project Period (FY): 2018-04-01 – 2020-03-31
• Keywords: 感覚予測誤差 / 報酬予測誤差 / 運動学習 / 行動選択

Outline of Final Research Achievements

In various situations, we decide on an action from several alternatives and execute a movement. If we don't get the expected result, the brain calculates two prediction errors, the sensory prediction error and the reward prediction error. Previous research has shown that reward prediction error is for appropriate action selection and sensory prediction error is for appropriate movement execution, but it is not fully known how these two errors interact. We performed behavioral experiments in which participants were exposed to unconscious visuomotor perturbation during target (reward) foraging tasks. We provided the sensory and reward information independently so that we desiccate that the effects of them on their exploratory movements. Surprisingly, we found that the movements were so volatile and directed into the large sensory prediction error when targets were applied with strong bias. This result imply that these two information were interested each other.

Free Research Field

身体教育学

Academic Significance and Societal Importance of the Research Achievements

従来の研究では、感覚予測誤差が生じるということは、運動を失敗させる要因になる排除すべきもの、と暗に捉えられてきた。しかし、本実験結果、脳は感覚予測誤差を報酬情報と捉えている可能性が見出されている。この結果は、これまでの定説を転回させる大きなインパクトを与えると期待できる。
また、従来の研究では、脳の報酬系と運動系は異なる領域で独立に存在すると考えられてきたが、感覚予測誤差によって行動選択が変調されることが明らかになれば、両者は深く影響を及ぼしあっていることが示唆される。これは、両者を繋ぐ脳内ネットワークに機能的意義を与えることを意味し、神経生理学研究などの関連分野への影響も大きい。