Hidden State Inference or Continuous Belief Updating during a Dynamic Visuomotor Skill.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Human Motor Control

Background:

• Adaptive sensorimotor behavior relies on updating beliefs with environmental changes.
• Predictive processing theories suggest hierarchical inference drives this flexibility.
• The study investigates continuous belief updating versus discrete state inference in action.

Purpose of the Study:

• To determine if continuous belief updating or discrete state inference better explains behavior in an interceptive task.
• To model perceptual inference using Bayesian and reinforcement learning frameworks.
• To compare the predictive power of different computational models on human behavior.

Main Methods:

• A virtual reality-based interceptive task with probabilistic cues and periodic context reversals.
• Recording participant gaze location prior to ball bounce as an index of perceptual inference.
• Applying Bayesian (Hierarchical Gaussian Filter) and associative learning models to analyze behavior.

Main Results:

• Continuous belief updating Bayesian models outperformed associative learning models in predicting behavior.
• Continuous belief updating better explained gaze behavior than discrete state inference.
• Despite better anticipating task structure, a state-inference model was outperformed.

Conclusions:

• Human sensorimotor behavior is characterized by continuous Bayesian belief updating.
• Behavior favors flexible, real-time updating over explicit structural inference in dynamic tasks.
• Cognitive processes show a gap between normative optimality and real-world adaptive action.