Bayesian methods for addressing long-standing problems in associative learning: The case of PREE

Area of Science:

• Cognitive Science
• Behavioral Neuroscience
• Computational Psychology

Background:

• Traditional associative learning models often assume single-state learning, focusing on gradual changes in associative strength.
• Evidence suggests multiple-state learning, where distinct relationships with shared features are learned independently, allowing rapid switching between behaviors.
• Theoretically, multiple-state learning requires distinct data analysis approaches, which have been under-explored.

Purpose of the Study:

• To present a Bayesian model capable of testing the predictions of the multiple-state learning view.
• To estimate the precise moment of behavioral state changes during acquisition and extinction phases.
• To demonstrate the utility of Bayesian methods in advancing associative learning research.

Main Methods:

• Developed a Bayesian model to analyze the Partial Reinforcement Extinction Effect (PREE).
• Applied the model to data from an experiment involving three reinforcement schedules (100%, 75%, 50%) during acquisition.
• Estimated the timing of response shifts from acquisition to extinction performance at individual and group levels.

Main Results:

• Identified significant differences in the estimated moment of switching between learned states during extinction.
• Observed that leaner partial reinforcement schedules during acquisition delayed the switch between behavioral states.
• Results provide empirical support for the multiple-state learning hypothesis.

Conclusions:

• The study provides the first direct test of predictions derived from the multiple-state learning view.
• Findings suggest that the degree of reinforcement influences the flexibility of switching between learned associations.
• Highlights the potential of Bayesian modeling for analyzing complex learning phenomena and advancing associative learning theory.