A new model of decision processing in instrumental learning tasks.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Science

Background:

• Decision-making and learning are interactive but often modeled separately.
• Existing joint models (RL-EAMs) using the diffusion decision model (DDM) fail to capture response time data in reinforcement learning.
• The DDM is limited to binary choices and struggles with absolute value effects.

Purpose of the Study:

• To develop a novel cognitive model that integrates reinforcement learning and evidence accumulation for improved decision-making analysis.
• To address the limitations of existing RL-EAMs, particularly the DDM, in capturing response time dynamics.
• To create a flexible framework applicable to multi-option choices and complex decision phenomena.

Main Methods:

• Proposed a new reinforcement learning-evidence accumulation model (RL-EAM) based on the advantage racing diffusion (ARD) framework.
• Extended the model to handle choices among two or more options.
• Evaluated the model's ability to capture response times, stimulus difficulty, speed-accuracy trade-offs, and reversal learning effects.

Main Results:

• The proposed RL-ARD model successfully captures crucial aspects of response times during reinforcement learning.
• The model demonstrates superiority over DDM-based approaches in explaining observed decision-making data.
• RL-ARD effectively models stimulus difficulty, speed-accuracy trade-offs, and reversal learning.

Conclusions:

• The RL-ARD framework offers a more accurate and comprehensive approach to modeling interactive learning and decision-making.
• This model overcomes fundamental limitations of the DDM, enabling analysis beyond binary choices.
• The computationally tractable RL-ARD provides a basis for broader applications in cognitive modeling and neuroscience.