A double error dynamic asymptote model of associative learning

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Learning Theory

Background:

• Traditional learning theories struggle to unify diverse associative learning phenomena.
• Existing models lack integrated representational and computational mechanisms for comprehensive prediction.

Purpose of the Study:

• To introduce a formal model of associative learning with enhanced predictive capabilities.
• To provide a unified account for phenomena that have eluded previous learning theories.

Main Methods:

• Developed the Double Error Dynamic Asymptote (DDA) model.
• Incorporated a fully connected network with temporally clustered stimuli.
• Introduced a "double error" term and a revaluation associability rate.
• Included a biologically plausible variable asymptote based on Hebbian learning principles.

Main Results:

• Simulations demonstrate the DDA model's ability to predict a wide range of associative learning phenomena.
• The model successfully integrates neutral stimuli associations and mediated learning.
• The DDA model accounts for reduced learning rates for expected predictors and attention shifts based on outcome predictiveness.

Conclusions:

• The DDA model offers a coherent framework for understanding associative learning.
• Its representational and computational mechanisms provide accurate predictions for complex learning behaviors.
• The model's biologically plausible components suggest potential neural underpinnings for associative learning.