An Information-Theoretic Framework for Understanding Learning and Choice Under Uncertainty.

Area of Science:

• Computational neuroscience
• Behavioral economics
• Information theory

Background:

• Information theory is extensively used in neuroscience for neural activity analysis.
• Its application to behavioral data, particularly discrete variables like choice and reward, is less explored.
• Discrete behavioral data is well-suited for information-theoretic analysis.

Purpose of the Study:

• To provide a framework using information theory for analyzing decision-making and learning strategies under uncertainty.
• To demonstrate how behavioral metrics can infer underlying cognitive mechanisms.
• To highlight information theory as a versatile, parameter-free tool for cognitive tasks.

Main Methods:

• Utilized simulated reinforcement-learning models as ground truth.
• Applied information-theoretic metrics, including conditional entropy and mutual information.
• Analyzed discrete, trial-by-trial behavioral data.

Main Results:

• Identified a positivity bias in learning rates (higher for rewards).
• Detected history-dependent changes in learning rates, indicating metaplasticity.
• Revealed reward harvest rate-driven adjustments in choice strategies and alternative learning strategies.

Conclusions:

• Information theory provides a powerful, parameter-free framework for analyzing complex behavioral strategies.
• This approach can uncover nuanced aspects of learning and decision-making under uncertainty.
• The framework has potential for broader application in cognitive science and neuroscience.