Kanji Shimomura

This paper develops a computational model of cortico-basal ganglia pathways with dual systems learning from positive and negative prediction errors, showing how opponent learning can lead to an obsession-compulsion cycle in OCD.

The paper explores how combining different state representations with appetitive and aversive learning improves performance in reward learning tasks, offering insights into cortico-basal ganglia circuits.

The paper proposes a dopamine-related mechanism involving successor representation and spiral RPE circuits that may explain difficulty in stopping habitual reward-seeking behaviors.