Related Experiment Videos
Base-rate effects in category learning: a comparison of parallel network and memory storage-retrieval models
W K Estes1, J A Campbell, N Hatsopoulos
1Department of Psychology, Harvard University, Cambridge, Massachusetts 02138.
Summary
Adaptive network models accurately predict category learning, outperforming exemplar models in capturing learning dynamics and base rate effects. Human performance closely matched predictions during learning but showed base-rate neglect in some test conditions.
Area of Science:
- Cognitive Psychology
- Computational Neuroscience
- Machine Learning in Psychology
Background:
- Category learning is fundamental to cognition, involving the organization of information into meaningful groups.
- Understanding how humans learn categories informs models of decision-making and pattern recognition.
- Previous models, like exemplar-memory, have limitations in explaining complex learning phenomena.
Purpose of the Study:
- To compare the predictive accuracy of adaptive network models versus exemplar-memory models for category learning.
- To investigate the influence of base rate effects on both learning and transfer performance in category tasks.
- To evaluate model performance against human subject data under varying feedback conditions.
Main Methods:
- Subjects classified patient symptom charts into disease categories.
- Feedback on classification accuracy was provided during learning trials.
- Feedback was manipulated (provided or withheld) during subsequent test trials to assess transfer.
Main Results:
- The adaptive network model demonstrated superior accuracy in modeling the detailed progression of category learning compared to the exemplar model.
- Both models predicted near-optimal use of category base rates and feature probabilities during learning, with the network model showing higher accuracy.
- Human subjects exhibited significant base-rate neglect under specific test conditions, a phenomenon partially predicted by the models.
Conclusions:
- Adaptive network models, due to their parallel processing and learning algorithms, offer a more robust framework for understanding category learning.
- Human category learning is sensitive to base rate information, but can deviate towards base-rate neglect, particularly in transfer or test phases.
- The findings highlight the strengths of network models in capturing the nuances of human learning and decision-making processes.