A reinforcement learning and sequential sampling model constrained by gaze data.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Science

Background:

• Reinforcement learning (RL) and sequential sampling models (SSMs) are used to analyze choice reaction time (RT) data.
• RL-SSMs effectively capture choice-RT patterns in repeated decision tasks.
• Integrating eye gaze data with RL-SSMs offers a novel approach to enhance predictive accuracy.

Purpose of the Study:

• To develop and evaluate a constrained reinforcement learning sequential sampling model (RL-SSM) incorporating eye gaze data.
• To investigate how learned option values and visual attention jointly influence evidence accumulation and choice.
• To bridge the gap between RL and visual attention modeling traditions.

Main Methods:

• Developed a novel computational model integrating RL-SSMs with eye-tracking data.
• Evaluated model performance on data from two eye-tracking experiments (N=133).
• Tested model variants with different value-gaze integration mechanisms.

Main Results:

• The integrated model significantly enhanced predictive ability compared to standard RL-SSMs.
• The model successfully captured empirical effects, including gaze biases on choice and response time.
• Demonstrated individual differences in valuation strategies (absolute vs. relative).

Conclusions:

• Constraining RL-SSMs with eye gaze data improves the understanding of decision-making.
• Learned values and visual attention interact dynamically to influence choice behavior.
• The unified model provides a powerful tool for studying the interplay of valuation and attention in choice.