Locally Bayesian learning with applications to retrospective revaluation and highlighting

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• Traditional associative learning models struggle to replicate complex human behavioral phenomena.
• Existing Bayesian learning models face challenges in capturing trial-order effects and retrospective revaluation.

Purpose of the Study:

• To develop a locally Bayesian parameter updating scheme for layered models.
• To enhance associative learning models' ability to capture human behavior, including retrospective revaluation and trial-order effects.

Main Methods:

• Implemented a scheme for locally Bayesian parameter updating in models with component functions.
• Utilized back-propagation of target data to interior modules for maximizing probabilistic targets.
• Applied the method to an associative learning model with attention filters.

Main Results:

• The model successfully exhibited retrospective revaluation effects like backward blocking and unovershadowing.
• The back-propagation of target values allowed the model to demonstrate trial-order effects, including highlighting.
• The approach captured differences in the magnitude of forward and backward blocking.

Conclusions:

• The locally Bayesian updating scheme offers a more effective way to model human associative learning.
• This method advances computational models by better explaining challenging behavioral phenomena.
• The approach provides a framework for developing more sophisticated and behaviorally accurate AI systems.