Episodic memory contributions to working memory-supported reinforcement learning.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Behavioral Economics

Background:

• Reinforcement learning (RL) models explain reward-based decision-making.
• Working memory (WM) and episodic memory (EM) also influence learning alongside RL.
• The specific contribution of EM to learning, particularly with WM and RL, remains unclear.

Purpose of the Study:

• To investigate the role of episodic memory (EM) in decision-making tasks where reinforcement learning (RL) and working memory (WM) are prominent.
• To develop novel experimental paradigms to isolate and quantify EM's contribution to learning.
• To understand how EM traces are encoded and retrieved during decision-making.

Main Methods:

• Two new experimental paradigms were designed to tag episodic memory (EM) using trial-unique stimuli and temporal context.
• Behavioral data were collected to assess learning and decision-making performance.
• Computational modeling was employed to analyze the contribution of EM traces to behavior.

Main Results:

• Episodic memory (EM) showed a consistently weak role in learning when combined with reinforcement learning (RL) and working memory (WM).
• EM did not enhance overall behavioral performance; instead, participants exhibited error patterns suggesting retrieval of stimulus-action choices without outcomes.
• Computational models confirmed a small contribution of EM traces of past stimulus-action events to learning.

Conclusions:

• Episodic memory (EM) has a limited impact on decision-making compared to reinforcement learning (RL) and working memory (WM).
• The way information is encoded and retrieved from EM may lead to characteristic errors in decision-making.
• Findings provide insights into the format of EM traces and their influence on cognitive processes supporting choice behavior.