Habit learning in hierarchical cortex-basal ganglia loops

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The interaction of multiple cortico-basal ganglia-thalamo-cortical (BGTC) loops remains a key question in understanding brain function.
• Existing models often consider BGTC loops as parallel, independent, or hierarchically organized, lacking a unified framework for their interaction in habit learning.

Purpose of the Study:

• To introduce and evaluate a neuro-computational model based on four key concepts to explain the interaction of BGTC loops in habit learning.
• To elucidate how different abstraction levels within BGTC loops contribute to action selection and learning.

Main Methods:

• Development of a neuro-computational model integrating four proposed key concepts.
• Evaluation of the model's computational benefits in simulating BGTC interactions.
• Comparison of the model's predictions with existing theories of habit learning.

Main Results:

• Each BGTC loop learns to select intermediate objectives at varying abstraction levels, from ventral striatum goals to putamen motor control.
• The cortex integrates BGTC selections with environmental feedback, while shortcuts between loops and individual prediction error signals are computed.
• The model demonstrates that collaborative loop function selects goal-directed actions, and slower shortcut training leads to habitual responses.

Conclusions:

• The proposed four key concepts provide a cohesive framework for understanding BGTC loop interactions in habit learning.
• This model highlights a collaborative mechanism for goal-directed actions and a distinct pathway for habitual response development.
• The findings offer new insights into the computational benefits and functional organization of BGTC circuits.