Integrated Bayesian models of learning and decision making for saccadic eye movements

Area of Science:

• Neuroscience
• Computational modeling
• Cognitive psychology

Background:

• Eye movement neurophysiology and decision-making models are well-studied.
• Linear rise-to-threshold models explain saccade latency variability but lack learning dynamics.
• Existing models do not account for conditional probabilities in sequential stimuli.

Purpose of the Study:

• To extend linear rise-to-threshold models to include learning across trials and conditional probabilities.
• To develop a hierarchical generative model integrating learning and decision-making processes.
• To enable statistical inference on learning and decision-making in eye movements.

Main Methods:

• Reformulated existing models into a hierarchical generative framework.
• Combined sub-models for trial-based learning and within-trial decision-making.
• Derived maximum-likelihood parameter estimation and model comparison using log likelihood ratios.

Main Results:

• Demonstrated the integrated model's utility on empirical saccade data from three subjects.
• Showed that eye movements reflect both marginal and conditional probabilities of target locations.
• Revealed distinct, subject-specific learning profiles identifiable by a naive Bayes classifier.

Conclusions:

• The extended model overcomes limitations of previous linear rise-to-threshold models.
• The approach allows for statistical inference on learning and decision-making in saccadic eye movements.
• Individual learning profiles are sufficiently distinct for subject identification.