How training and testing histories affect generalization: a test of simple neural networks

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Generalization experiments in associative learning reveal complex phenomena like errorless learning, peak shift, and central tendency.
• Existing models often struggle to account for these specific training and testing procedure-dependent findings.
• Neural networks offer a promising framework for modeling sequential experiences with stimuli.

Purpose of the Study:

• To test a simple network model of associative learning against specific generalization findings.
• To determine if basic associative memory mechanisms can explain complex stimulus range effects.
• To highlight the utility of neural networks in studying phenomena dependent on sequences of experiences.

Main Methods:

• Development of a simple network model for associative learning.
• Simulation of training and testing procedures common in generalization experiments.
• Comparison of model predictions with empirical data on errorless learning, peak shift, and central tendency effects.

Main Results:

• The network model successfully reproduced the effects of errorless learning.
• The model accurately predicted the impact of extinction testing on peak shift.
• The central tendency effect was also replicated by the network model.

Conclusions:

• A simple associative learning network model can account for specific findings in generalization.
• Complex phenomena like stimulus range effects may arise from fundamental associative memory processes.
• Neural networks provide a valuable tool for investigating experience-dependent learning phenomena.