Learning about the internal structure of categories through classification and feature inference

Area of Science:

• Cognitive Psychology
• Machine Learning
• Artificial Intelligence

Background:

• Category learning research suggests classification tasks create feature-skewed representations, while inference tasks yield richer within-category structures.
• Existing studies often overlook how typical and atypical features are distributed within a category, focusing only on typical features.

Purpose of the Study:

• To test if inference learning results in richer knowledge of internal category structure compared to classification learning.
• To introduce novel measures for probing learners' representations of within-category structure.

Main Methods:

• Two experiments were conducted to compare category knowledge representations.
• New measures were developed to assess the distribution of typical and atypical features within learned categories.

Main Results:

• Inference learners used a wider range of feature dimensions, but classification learners showed greater sensitivity to atypical features.
• Classification learners were more likely to incorporate atypical features into their representations, even when attention was directed to classification-relevant dimensions.

Conclusions:

• Results contradict the hypothesis that inference learning yields superior knowledge of within-category structure.
• Classification learning enhances sensitivity to the distribution of both typical and atypical features, offering a more nuanced understanding of category structure.