Neural network modeling of developmental effects in discrimination shifts
1McGill University, Montréal, Quebec, Canada. sirois@psych.mcgill.ca
Journal of Experimental Child Psychology
|January 8, 1999
Summary
Neural networks effectively simulate developmental changes in discrimination shift learning. Increased training depth in these networks mimics human developmental phenomena, suggesting overtraining influences learning differences.
Area of Science:
- Cognitive Science
- Computational Neuroscience
- Developmental Psychology
Background:
- Discrimination shift learning involves understanding empirical regularities.
- Existing psychological theories lack a comprehensive account of shift learning development.
- Previous neural network simulations of shift learning have been unsuccessful.
Purpose of the Study:
- To simulate developmental phenomena in discrimination shifts using neural networks.
- To evaluate the efficacy of the cascade-correlation algorithm in modeling shift learning.
- To propose an explanation for developmental differences in shift learning.
Main Methods:
- Review of discrimination shift literature and theoretical accounts.
- Implementation of neural network simulations using the cascade-correlation algorithm.
- Manipulation of network training duration to simulate developmental changes.
Main Results:
- Neural networks using cascade-correlation captured empirical regularities of discrimination shifts effectively.
- Adjusting training depth in networks simulated developmental phenomena observed in human learning.
- The model demonstrated superior performance compared to existing psychological theories.
Conclusions:
- Neural networks can model discrimination shift learning development more accurately than current theories.
- Simulated overtraining in networks provides a plausible explanation for developmental differences in human shift learning.
- Findings align with existing literature on overtraining and learning.
