Ventral striatum uses a temporal difference rule for prediction during neuroprosthetic control

摘要

During neuroprosthetic control, prediction of future outcome plays a key role, by informing about current performance, so the subject can correct and refine its actions. During performance, an error signal emerges representing the difference between what was predicted and what actually happened. Several studies have focused on understanding how error signals emerges. Yet, the predictions needed to compute these error signals are still unclear. It has been suggested that the ventral striatum, may be the responsible for encoding a prediction function. So far, there is no clear evidence that these predictions are present in the ventral striatum or which different roles dorsal and the ventral striatum may be playing during neuroprosthetic control. To that effect, we trained rats to control a brain-machine-interface (BMI) with motor cortex units while recording the activity in striatum. Our results show that neuronal activity of the ventral striatum could predict future outcome and that this prediction coincided with the state prediction of a temporal difference model of reinforcement learning. These results highlight the relevance of the ventral striatum during neuroprosthetic control and point towards the use of neuronal prediction information to improve brain-machine-interfaces.