An adaptive linear filter model of procedural category learning

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Machine Learning

Background:

• Category learning is fundamental to cognition.
• Existing models struggle to explain individual and group learning dynamics.
• Procedural categorization models require further development.

Purpose of the Study:

• To introduce and validate an Adaptive Linear Filter (ALF) model for category learning and transfer.
• To assess the ALF model's performance against empirical data at both group and individual levels.
• To demonstrate the ALF model's advantages over alternative associative and prototype models.

Main Methods:

• Utilized a feature-based association model (ALF) with a logistic output function and Least Mean Squares learning.
• Applied the ALF model to 31 diverse published datasets covering category learning and transfer.
• Performed both grouped-level and individual-level data analyses.

Main Results:

• The ALF model demonstrated remarkable performance, accounting for substantial variance in both grouped and individual data.
• The model outperformed alternative models when fitted to grouped data.
• High explained variances were achieved for individual learning and transfer performance with minimal free parameters.

Conclusions:

• The Adaptive Linear Filter (ALF) model provides a parsimonious and effective account of procedural categorization.
• The ALF model successfully captures empirical trends and addresses limitations of other associative models.
• The study clarifies that the ALF model is distinct from prototype models in its mechanism.