The effect of training methodology on knowledge representation in categorization

Area of Science:

• Cognitive Science
• Computational Psychology
• Machine Learning

Background:

• Category representations influence decision-making, but factors affecting their development and generalizability are unclear.
• Distinction between within-category and between-category information is crucial for understanding cognitive processes.
• Investigating training methodology and category structures is essential for advancing categorization research.

Purpose of the Study:

• To investigate how training methods and category structures influence the development of within-category versus between-category representations.
• To examine the generalizability of different category representations using computational modeling.
• To determine the dominant and robust representation type for knowledge reconfiguration.

Main Methods:

• Employed traditional empirical methods and computational modeling from machine learning.
• Experiment 1 used rule-based (RB) category structures with classification and concept training.
• Experiment 2 used information-integration (II) category structures to assess representation biases.

Main Results:

• Classification training favored between-category representations in RB structures, while concept training favored within-category representations.
• Information-integration structures consistently promoted within-category representations, irrespective of training method.
• Computational modeling indicated that only within-category representations supported generalization in both experiments.

Conclusions:

• Within-category representations appear dominant and more robust for supporting generalization.
• Training methodology interacts with category structure to shape representations, particularly in rule-based tasks.
• Findings highlight the importance of within-category representations for flexible knowledge application and generalization.