Building and Breaking the Chain: A Model of Reward Prediction Error Integration and Segmentation of Memory

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Modeling

Background:

• Prediction errors are crucial for reinforcement learning and episodic memory organization.
• Midbrain dopamine, locus coeruleus, and hippocampus play key roles in event cognition.
• The interaction between prediction errors in memory integration and segmentation remains unclear.

Purpose of the Study:

• To propose and simulate the theoretical influence of two prediction error signals on event integration and segmentation in memory.
• To investigate how signed and unsigned reward prediction errors differentially affect mental models and contextual similarity.
• To explain the impact of contextual similarity on temporal-order memory patterns.

Main Methods:

• Review of neuromodulatory systems (dopamine, locus coeruleus, hippocampus) in event cognition.
• Development of a computational model simulating prediction error signals.
• Analysis of model performance in associative chaining and sequential retrieval tasks.

Main Results:

• Signed reward prediction errors promote contextual similarity (integration) by building stable reward environment models.
• Unsigned reward prediction errors signal new environmental models, causing contextual shifts (segmentation).
• Simulations demonstrated differential effects of integration and segmentation on associative chaining and temporal-order memory.

Conclusions:

• A unifying explanation for how neuromodulatory systems influence event cognition and memory through prediction errors.
• Differential roles of signed (integration) and unsigned (segmentation) prediction errors in memory organization.
• Computational model successfully accounts for complex temporal-order memory phenomena.