Cognitive control over learning: creating, clustering, and generalizing task-set structure.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Learning and executive functions, like task-switching, share neural bases in the prefrontal cortex and basal ganglia.
• Understanding their interaction is key: cognitive control aids learning, while learning provides structure for control.

Purpose of the Study:

• Investigate how cognitive control and learning interact to create structured representations.
• Model the inference of hidden structures and the reuse/creation of task-sets.
• Link computational models to neurobiology and human behavior.

Main Methods:

• Developed a context-task-set (C-TS) model using nonparametric Bayesian methods.
• Created a neurobiologically explicit network model of frontal cortex and basal ganglia circuits.
• Tested model predictions against human choices and response times during learning and task-switching.

Main Results:

• The network model approximates C-TS computations, with neural mechanisms modulating C-TS parameters.
• Models predict spontaneous task-set structuring by participants, leading to positive/negative transfer.
• Experimental data support C-TS model predictions for human choice sequences.

Conclusions:

• Cognitive control and learning interact dynamically, forming abstract representations.
• This structured learning supports generalization and long-term optimality over short-term gains.
• Findings highlight the brain's tendency to impose structure for enhanced learning and transfer.