Array training in a categorization task.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience

Background:

• Categorization relies on identifying within-category commonalities and between-category distinctiveness.
• Traditional categorization studies often use single-pattern presentations.

Purpose of the Study:

• To investigate how array training, versus single-pattern presentation, affects categorization learning and transfer.
• To compare the predictive power of the generalized context model and a prototype model in explaining categorization performance.

Main Methods:

• Subjects learned three prototype categories using either array training (same or different categories) or single-pattern presentation.
• Performance was assessed on old and new patterns across low, medium, and high distortion levels.
• Computational models (generalized context model, prototype model, exemplar model) were used to predict results.

Main Results:

• Array training enhanced learning, particularly when presented with patterns from the same category.
• A strong gradient effect was observed across distortion levels, with highest performance after array training on different category patterns.
• Neither the generalized context model nor the exemplar model fully predicted the observed gradient and pattern of results; a prototype model offered a better fit.

Conclusions:

• The way category information is presented during learning significantly impacts categorization performance.
• Prototype models provide a better account of categorization data than the generalized context model in this paradigm.
• Both within-category commonalities and between-category distinctiveness play a crucial role in forming categorical representations.