[Multiple Dopamine Signals and Their Contributions to Reinforcement Learning]

Masayuki Matsumoto
DOI: https://doi.org/10.11477/mf.1416200567
Abstract:Midbrain dopamine neurons are activated by reward and sensory cue that predicts reward. Their responses resemble reward prediction error that indicates the discrepancy between obtained and expected reward values, which has been thought to play an important role as a teaching signal in reinforcement learning. Indeed, pharmacological blockade of dopamine transmission interferes with reinforcement learning. Recent studies reported, however, that not all dopamine neurons transmit the reward-related signal. They found that a subset of dopamine neurons transmits signals related to non-rewarding, salient experiences such as aversive stimulations and cognitively demanding events. How these signals contribute to animal behavior is not yet well understood. This article reviews recent findings on dopamine signals related to rewarding and non-rewarding experiences, and discusses their contributions to reinforcement learning.
What problem does this paper attempt to address?