Toward a common representational framework for adaptation.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Machine Learning

Background:

• Instance theory and strength theory offer distinct explanations for learning dynamics.
• Reconciling these theories computationally is crucial for understanding adaptive behavior.

Purpose of the Study:

• To develop a computational model (ARM) unifying instance and strength learning theories.
• To investigate mechanisms driving adaptive behavior in dynamic environments using ARM.
• To test model variants against experimental data on learning shifts.

Main Methods:

• Developed the adaptive representation model (ARM) to integrate instance and strength theories.
• Created 10 ARM variants to instantiate different adaptive mechanisms.
• Fit models to single-trial choice response time data from three experiments with varying category distributions.

Main Results:

• Showed instance representations as a special case of strength-based representations within ARM.
• Identified prediction error and lateral inhibition as key mechanisms for adaptive behavior.
• Demonstrated that choice and response time distributions emerge from evolving stimulus-response mappings.

Conclusions:

• ARM provides a unified framework for studying learning dynamics.
• Prediction error and lateral inhibition are vital for adapting to changing environments.
• The model successfully captures emergent choice and response time patterns.