|
To decide : the role of dopamine in planning and action| old_uid | 4 |
|---|
| title | To decide : the role of dopamine in planning and action |
|---|
| start_date | 2005/09/12 |
|---|
| schedule | 11h30 |
|---|
| online | no |
|---|
| location_info | / |
|---|
| details | invitée par Thomas Boraud |
|---|
| summary | According to current models of the basal-ganglia as the neural substrate underlying reinforcement learning, dopamine neurons provide a reinforcement signal reporting the change in prediction of reward.
In the first study we report, we recorded midbrain dopamine neurons in monkeys performing a probabilistic instrumental conditioning task. We found that dopamine neurons accurately report the mismatch between the monkeys’ expectations and reality, closely resembling the TD error signal. Yet, this study, and other physiological studies are limited to classical conditioning (CC) and instrumental conditioning (IC), avoiding the question of decision-making. We therefore examined behaviour and dopamine activity in a 2-armed bandit task vs. a reference IC task involving the same stimuli.
The place of dopamine in decision behaviour was explored on two time-scales. First, we show that while the monkeys’ long term decision policy follows models of reward reinforcement, behaviour is best explained by its neural correlate, the dopamine responses in the reference task. Secondly, we demonstrate that in the decision task, future action is already encoded in the dopamine responses when the stimuli are presented. This locates the dopamine neurons upstream or at the centre of the decision junction. We further infer that the effect of dopamine on behaviour is one of long-term reinforcement, rather than immediate action selection.
Finally, our results call for an update of computational basal-ganglia models to implement learning of state-action values, rather than state alone. |
|---|
| responsibles | Renaud, Deris |
|---|
| |
|