The Influence of Presentation Order on Category Transfer

Area of Science:

• Cognitive Psychology
• Machine Learning
• Computational Neuroscience

Background:

• Category learning is fundamental to cognition.
• Understanding how training data influences learning is crucial for artificial intelligence and human learning.
• Previous research suggests different training methods yield distinct category representations.

Purpose of the Study:

• To investigate how the order of training examples affects supervised categorization.
• To determine if rule-based or similarity-based training leads to different category knowledge.
• To examine the transfer of category learning to new stimuli based on presentation order.

Main Methods:

• Utilized the 5-4 category structure from Medin and Schaffer (1978).
• Implemented two training conditions: rule-based (grouped by subcategories, randomized within) and similarity-based (grouped by featural similarity, ignoring subcategories).
• Assessed transfer of category learning to new stimuli.

Main Results:

• Transfer patterns were dependent on the training presentation order (similarity-based vs. rule-based).
• Participants particularly leveraged the rule-based order for category learning.
• Rule-based training facilitated abstract logical rule learning and rule-based classification.

Conclusions:

• The order of training examples is a critical factor in supervised categorization.
• Rule-based presentation orders promote the development of abstract rule-based category knowledge.
• Findings have implications for designing effective training paradigms in both human and machine learning.