Hebbian Learning in a Random Network Captures Selectivity Properties of the Prefrontal Cortex.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Neuroscience

Background:

• Complex cognitive functions like context-switching rely on the prefrontal cortex (PFC).
• Neural activity in the PFC requires task specialization and flexibility.
• Nonlinear "mixed" selectivity is a key neurophysiological trait for complex, context-dependent behaviors.

Purpose of the Study:

• Investigate the extent of computationally relevant properties, such as mixed selectivity, in the PFC.
• Explore how circuit mechanisms could give rise to these properties.

Main Methods:

• Recorded neural activity from PFC cells in rhesus macaques during a complex task.
• Developed computational models with random feedforward inputs and Hebbian learning rules.
• Analyzed neural data and model outputs for properties like mixed selectivity, noise, response density, and clustering.

Main Results:

• PFC cells showed moderate specialization and structure, exceeding predictions from a random input model.
• A simple Hebbian learning rule significantly increased mixed selectivity in the model, matching experimental data.
• The learned model accurately reproduced noise, response density, clustering, and selectivity distributions observed in the data.

Conclusions:

• Hebbian learning is a plausible mechanism for generating the mixed selectivity observed in the PFC.
• This learning process refines neural circuits to support complex cognitive behaviors.
• The study provides testable predictions for how selectivity measures evolve during animal training.