Stimulus generalization in two associative learning processes

Area of Science:

• Cognitive Psychology
• Human Associative Learning
• Machine Learning Models

Background:

• Human associative learning research presents conflicting findings regarding stimulus generalization.
• Nonlinear discrimination tasks suggest narrow generalization, where similar stimuli weakly activate representations.
• Cue-interaction phenomena indicate broad generalization in associative learning.

Purpose of the Study:

• To investigate the limitations of current associative learning models in explaining both narrow and broad generalization.
• To determine if a single process can account for nonlinear discrimination and cue interaction.
• To propose a dual-process model for human associative learning.

Main Methods:

• Conducted three experiments examining human associative learning.
• Analyzed data in the context of existing computational models of associative learning.
• Compared model predictions with empirical findings on discrimination and cue interaction.

Main Results:

• Current associative learning models fail to simultaneously explain narrow generalization (discrimination) and broad generalization (cue interaction).
• Evidence suggests that nonlinear discrimination is driven by configural stimulus representation with narrow generalization.
• Cue interaction is better explained by an adaptive learning process with broad generalization.

Conclusions:

• Human associative learning likely involves at least two distinct processes.
• An exemplar-based process with configural stimulus representation underlies narrow generalization.
• A separate adaptive learning process accounts for broad generalization and cue interaction.