Belief inference for hierarchical hidden states in spatial navigation.

Area of Science:

• Cognitive Neuroscience
• Computational Psychiatry
• Decision Making

Background:

• Real-world decision-making often involves navigating environments with multiple unobservable hidden states.
• Resolving uncertainty through mutual inference is crucial for effective decision-making in complex environments.

Purpose of the Study:

• To develop and analyze a spatial navigation task (Tiger maze) requiring simultaneous inference of local and global hidden states from uncertain observations.
• To utilize a Bayesian computational approach with a hierarchical inference model to understand human behavior in this task.
• To investigate the neural correlates of belief reassessment and reinforcement learning during hidden state inference using functional magnetic resonance imaging (fMRI).

Main Methods:

• Development of a novel 'Tiger maze' task for spatial navigation under uncertainty.
• Application of a hierarchical Bayesian inference model to human behavioral data.
• Utilizing functional magnetic resonance imaging (fMRI) to capture neural activity during task performance.

Main Results:

• The study successfully separated neural correlates of reinforcement learning from belief reassessment in hidden states.
• fMRI data indicated differential involvement of the basal ganglia and dorsomedial prefrontal cortex based on uncertainty layers.
• Neural regions were organized along a rostral axis, correlating with inference type and hidden state abstraction level (higher-order inference engaging more anterior regions).

Conclusions:

• The hierarchical Bayesian model provides a computational framework for understanding uncertainty resolution in spatial navigation.
• Distinct neural circuits, organized rostrocaudally, support different levels of inference and abstraction in hidden states.
• This research elucidates the neural basis of complex decision-making under uncertainty, with implications for understanding cognitive processes.