演者 内田 直滋 博士
Transcription
演者 内田 直滋 博士
日時 2015年3月2日(月)17:30-19:00 会場 東北大学 医学部1号館 1階 第一講義室 演題 Arithme9c of dopamine reward predic9on errors (ドーパミン・ニューロンは何を計算しているのか?) 演者 内田 直滋 博士 Professor, Center for Brain Science, Harvard University It has been proposed that dopamine neurons in the midbrain signal reward predic9on errors, that is, the discrepancy between actual and expected reward (Schultz et al., 1997; Bayer and Glimcher, 2005). Reward predic9on error = Actual reward -‐ expected reward These signals resemble error signals used to train computers in machine learning or ar9ficial intelligence. However, the mechanism underlying this calcula9on in the brain remains unknown. To probe how dopamine neurons calculate reward predic9on error, we have developed a mouse model that allows us to combine electrophysiology in behaving animals with emerging molecular and gene9c techniques (Cohen et al., 2012). In a recent study (Eshel et al., in prepara9on), we recorded from optogene9cally-‐iden9fied dopamine neurons in the ventral tegmental area (VTA) while mice performed classical condi9oning tasks that varied expecta9on level, reward size, or both. We found that a simple, universal func9on predicts how individual dopamine neurons will respond to unexpected rewards of various sizes. In the presence of expecta9on, this func9on shiXs downward in a purely subtrac9ve manner, consistent with the above, canonical predic9on error equa9on. Furthermore, the effect of expecta9on on each neuron’s reward response mul9plica9vely scaled with the responsiveness of that neuron. In other words, each dopamine neuron appears to calculate reward predic9on error in the same way, but scaled upwards or downwards. Such a process could naturally emerge from a homeosta9c balance of excita9on and inhibi9on, and allows each dopamine neuron to contribute fully to the predic9on error signal. ※ 本セミナーは医学系研究科系統講義コース科目の授業として振替可能なセミナーです。 連絡先:情報科学研究科 数学連携推進室 三浦 佳二 医学部 発生発達神経科学分野 大隅 典子(内線8203)