Predicting category intuitiveness with the rational model, the simplicity model, and the generalized context model

Area of Science:

• Cognitive Psychology
• Machine Learning
• Computational Neuroscience

Background:

• Category intuitiveness, or how naturally stimuli group, is usually studied with unsupervised learning models.
• Predicting intuitive category structures is crucial for understanding human cognition.

Purpose of the Study:

• To develop a novel measure of category intuitiveness using a supervised learning model.
• To evaluate the predictive power of this new approach against existing unsupervised models.

Main Methods:

• The generalized context model (GCM), a supervised learning model, was adapted to predict category assignments based on all other instances.
• This adapted GCM, termed 'unsupervised GCM,' was used to determine the most intuitive category assignment by minimizing prediction error.
• Predictions were compared against established unsupervised models: the simplicity and rational models.

Main Results:

• The unsupervised GCM demonstrated competitive performance, comparing favorably with both the simplicity and rational models.
• This suggests that supervised learning frameworks can effectively model aspects of unsupervised category perception.
• No single model significantly outperformed all others, indicating ongoing research needs.

Conclusions:

• The distinction between supervised and unsupervised categorization may be less rigid than previously thought.
• The generalized context model offers a viable framework for exploring category intuitiveness.
• Further research is required to fully understand and model the nuances of category intuitiveness.