Entropy-based metrics for predicting choice behavior based on local response to reward.

Area of Science:

• Behavioral neuroscience
• Computational neuroscience
• Information theory

Background:

• The matching law has long quantified animal choice behavior based on reinforcement.
• Reinforcement learning (RL) models explain choice by integrating reward feedback over time.
• Current RL models struggle to account for variability in matching behavior.

Purpose of the Study:

• To develop novel metrics for quantifying choice variability.
• To improve the accuracy of computational models of animal choice.
• To explore the relationship between information theory and adaptive behavior.

Main Methods:

• Applied information theory metrics, specifically entropy-based measures, to choice data.
• Utilized data from dynamic learning tasks in mice and monkeys.
• Developed enhanced RL models informed by information-theoretic limitations.

Main Results:

• A single entropy-based metric explained 50% of matching variance in mice.
• The same metric explained 41% of matching variance in monkeys.
• New RL models incorporating entropy metrics demonstrated improved accuracy.

Conclusions:

• Entropy-based metrics provide a powerful model-free tool for predicting adaptive choice.
• These metrics can reveal underlying neural mechanisms of decision-making.
• Information theory offers a valuable framework for advancing computational models of behavior.