Testing models of context-dependent outcome encoding in reinforcement learning.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Behavioral Economics

Background:

• Reinforcement learning (RL) research indicates that the evaluation of choice outcomes is influenced by context.
• Existing computational models, such as reference point centering and range adaptation, attempt to explain this context-dependent encoding.

Purpose of the Study:

• To introduce and evaluate novel computational models for context-dependent outcome encoding in human RL.
• Specifically, to test a frequency encoding model and a range-frequency model against established theories.

Main Methods:

• Two fully incentivized behavioral experiments were designed with choice tasks.
• These tasks were crafted such that candidate computational models would make distinct predictions.

Main Results:

• Behavioral data analysis revealed that models incorporating frequency or rank-based encoding provided the best fit.
• Results were most consistent with the proposed frequency encoding and range-frequency models.

Conclusions:

• The study provides evidence for frequency and rank as key mechanisms in context-dependent outcome evaluation in human RL.
• Findings enhance the understanding of the computational processes underlying human decision-making in varying contexts.