Equivalence Projective Simulation as a Framework for Modeling Formation of Stimulus Equivalence Classes

Area of Science:

• Behavioral psychology
• Artificial intelligence
• Machine learning

Background:

• Stimulus equivalence (SE) and projective simulation (PS) are frameworks for studying complex behavior in humans and artificial agents, respectively.
• Existing models often lack interpretability and flexibility.
• Bridging behavior analysis and machine learning is a significant challenge.

Discussion:

• The equivalence projective simulation (EPS) model is proposed, adapting the PS framework to model SE formation.
• EPS simulates the emergence of equivalence relations using a matching-to-sample (MTS) procedure.
• The model operates within a varying action set and derives new relations without environmental feedback.

Key Insights:

• This research marks the first link between equivalence theory in behavior analysis and artificial agents within machine learning.
• The EPS model demonstrates advantages over traditional neural network models due to its interpretability and flexibility.
• The model successfully replicates experimental results, including typical and atypical participant behavior and the nodal effect.

Outlook:

• The EPS model provides a valuable tool for understanding and simulating complex cognitive processes.
• Further research can explore the application of EPS in diverse areas of equivalence theory.
• The model's interpretability facilitates modifications for advanced research questions in artificial intelligence and cognitive science.