The binding pool: a model of shared neural resources for distinct items in visual working memory

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Psychology

Background:

• Visual working memory (VWM) theories propose discrete slots or shared resources for information storage.
• Neural mechanisms underlying VWM theories remain poorly understood.
• Existing models lack a unified neural account for VWM storage and retrieval.

Purpose of the Study:

• To propose a computationally realized neural account for VWM.
• To integrate discrete slot and shared resource theories within a unified model.
• To explain the neural basis of storing and retrieving multiple visual stimuli.

Main Methods:

• Developed the binding pool model, a neural network simulating VWM.
• The model uses a type/token architecture with shared neural resources.
• Simulated data from continuous report and change detection tasks.

Main Results:

• The binding pool model stores distinct objects with bound features in shared resources.
• Increased memory load leads to mutual interference between stored items.
• The model generates predictions on accuracy, confidence, and stimulus similarity interactions.

Conclusions:

• The binding pool model offers a hybrid approach to VWM, merging slot and resource concepts.
• It provides a neural basis for understanding interference in VWM.
• Testing model predictions will clarify the role of shared resources and neural ensembles in visual memory.