Given versus induced category representations: use of prototype and exemplar information in classification

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Machine Learning

Background:

• Category acquisition can be induced through experience with exemplars or given via direct prototype presentation.
• Understanding how these information types interact is crucial for cognitive models of learning.
• A relational coding model integrating exemplar and prototype information was proposed.

Purpose of the Study:

• To investigate the effects of induced versus given category information on classification.
• To test a mixture model accounting for categorization based on both prototype and exemplar data.
• To differentiate classification strategies based on training methods.

Main Methods:

• Two experiments were conducted using geometric shapes with four binary dimensions for two ill-defined categories.
• Subjects were trained in three conditions: exemplars only, prototypes then exemplars, or concurrent prototypes and exemplars.
• Classification tests assessed performance on prototypes, old exemplars, and new exemplars.

Main Results:

• A mixture model accurately predicted category judgments across all training groups by varying prototype and exemplar information use.
• Subjects receiving direct prototype information used a mix of prototype and exemplar similarity.
• Subjects trained only on exemplars relied solely on exemplar similarity, despite accurate prototype judgment.

Conclusions:

• The type of category information (induced vs. given) significantly impacts classification strategies.
• Learners integrate prototype and exemplar information differently based on training modality.
• The relational coding model effectively captures the interplay between abstracted category-level information and exemplar-based representations.