Reinforcement learning can account for associative and perceptual learning on a visual-decision task

Area of Science:

• Neuroscience
• Computational Neuroscience
• Perceptual Learning

Background:

• Perceptual performance improvements are linked to altered interpretation of sensory information for decision-making.
• Understanding the neural mechanisms underlying these interpretive changes is crucial for explaining learning.

Purpose of the Study:

• To investigate if a reinforcement-learning rule can explain changes in functional connectivity driving perceptual learning.
• To model how neural connections adapt to improve performance in a visual motion direction-discrimination task.

Main Methods:

• Simulated responses from direction-selective sensory neurons in the middle temporal area (MT) of monkey cortex were used.
• A computational model incorporated a reinforcement-learning rule, utilizing reward prediction error to modify synaptic connections.
• The model assessed how changes in connectivity between sensory and decision neurons affected behavioral performance.

Main Results:

• A reinforcement-learning mechanism successfully accounted for performance improvements in the perceptual task.
• Reward prediction errors guided the strengthening of connections from sensory to decision neurons.
• Selective strengthening of connections from highly sensitive sensory neurons gradually enhanced perceptual sensitivity.

Conclusions:

• Functional connectivity between sensory and decision neurons can be shaped by reinforcement learning.
• This mechanism provides a unified explanation for both associative and perceptual learning.
• Feedback-driven processes play a significant role in adapting neural circuits for improved perception and behavior.