A shared predictive architecture in the sensory cortex for statistical and reward-based learning.

Area of Science:

• Neuroscience
• Sensory Processing
• Learning and Memory

Background:

• The sensory cortex is traditionally viewed as a feed-forward feature extractor.
• Emerging evidence suggests a more complex role, including prediction error computation and reward-based prediction.
• This challenges the traditional view by highlighting the cortex's active role in prediction.

Purpose of the Study:

• To demonstrate that the sensory cortex has a core function in prediction, beyond feature representation.
• To propose a circuit motif for implementing predictive functions within the sensory cortex.
• To review empirical evidence supporting this predictive role, primarily from the auditory cortex.

Main Methods:

• Review of existing empirical evidence from auditory and other sensory cortices.
• Analysis of studies on implicit statistical learning and explicit reward-based learning.
• Theoretical proposal of a circuit motif involving dendritic inputs and local disinhibition.

Main Results:

• Sensory cortices exhibit prediction error signals during statistical learning.
• Sensory cortex populations rapidly develop reward prediction activity during reward-based learning.
• A specific circuit motif can theoretically implement both prediction error computation and simple prediction.

Conclusions:

• The sensory cortex plays a dual role: feature extraction and prediction.
• A unified circuit motif can explain how the sensory cortex computes prediction errors and makes predictions.
• Further research is needed to validate this principle across sensory systems.