A simple model of prefrontal cortex function in delayed-response tasks

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Neurobiology

Background:

• Prefrontal cortex (PFC) functions are crucial for cognitive tasks.
• Delayed response (DR), AB, and delayed matching-to-sample (DMS) tasks are standard tools for studying PFC.
• Understanding PFC mechanisms requires integrating behavioral and neural data.

Purpose of the Study:

• To develop a minimal neural network simulation model.
• To link behavioral and electrophysiological data from DR, AB, and DMS tasks.
• To elucidate the role of PFC in rule-based learning and behavior.

Main Methods:

• A two-level neural network architecture was designed, inspired by neuroanatomy.
• Level 1 simulated a sensory-motor loop; Level 2 incorporated memory and rule-coding units.
• The model was trained using visual objects and reinforcement signals, with and without Level 2.

Main Results:

• Simulating Level 1 alone resulted in errors similar to prefrontal lesions (e.g., AB error).
• Adding Level 2 enabled the network to learn task rules through reinforcement.
• The network demonstrated rapid adaptation to changing reinforcement schedules.

Conclusions:

• A two-level neural network can model PFC functions in tasks like DR, AB, and DMS.
• Rule acquisition and adaptation depend on higher-level processing units modulating sensory-motor loops.
• The model provides insights into PFC mechanisms, systematicity in neural networks, and 'mental Darwinism'.