"Neural circuit mechanism for the computation of reward prediction error in the midbrain dopamine neurons", Kenji Morita (University of Tokyo)
Seminar by Dr. Kenji Morita, 14:00PM Friday, August 24 @ D015-Lab1
Neural Computation Unit would like to invite you to a seminar by Dr. Kenji Morita from Physical and Health Education, Graduate School of Education, The University of Tokyo.
- Date: Friday, August 24
- Time: 14:00-
- Venue: Lab 1, Meeting room D015 (Level D)
- Title: Neural circuit mechanism for the computation of reward prediction error in the midbrain dopamine neurons
- Speaker: Dr. Kenji Morita, Physical and Health Education, Graduate School of Education, The University of Tokyo
Midbrain dopamine neurons supposedly encode reward prediction error, but how it is computed remains elusive. We propose a mechanism based on recent anatomical and physiological findings (Morita et al., 2012, Trends Neurosci 35:457): (1) two types of corticostriatal neurons, crossed corticostriatal (CCS) cells and corticopontine/pyramidal-tract
(CPn/PT) cells, represent the subject's current and previous state/ action, respectively, through unidirectional projections from CCS cells to CPn/PT cells (Morishima & Kawaguchi, 2006, J Neurosci 26:4394) and strong facilitating recurrent excitation only among CPn/PT cells (Morishima et al., 2011, J Neurosci 31:10380), and (2) since the CCS and CPn/PT cells predominantly target the striatal medium spiny neurons projecting to the basal ganglia direct and indirect pathways, respectively (Reiner et al., 2010, Front Neuroanat, 4:142), and those pathways supposedly up- and down-regulate dopamine neurons, the temporal difference of the state/action values, the core of reward prediction error, can be computed. This proposed mechanism predicts that the degree of time discount for future rewards is determined by the balance between the direct and indirect pathways, which can be modulated by serotonin operation in the globus pallidus and the cortex. We have also proposed, based on the argument that striosomes/patches in the striatum could be predominantly driven by the CPn/PT cells (see Crittenden & Graybiel, 2011, Front Neuroanat, 5:59 for the debate), a possibility that the striosomal projections to dopamine neurons also contribute to computing previous state/action values. Our hypothesis suggests a unified view of basal ganglia functions in reinforcement learning and motor control, and has important clinical implications.
We look forward to seeing you in the seminar.
Kenji Doya, Neural Computation Unit
posted by Emiko Asato on behalf