Modeling visual working memory using recurrent on-center off-surround neural network with distance dependent...

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Computer Vision

Background:

• Visual working memory (VWM) is crucial for retaining and manipulating visual information.
• Current models often follow a feedforward processing order, which may not fully capture VWM dynamics.
• Understanding the hierarchical nature of visual processing is key to modeling VWM.

Purpose of the Study:

• To present a novel hierarchical computational model of visual working memory (VWM).
• To simulate the processing of spatially distributed objects and their features with a reversed processing order.
• To investigate the impact of feedback mechanisms and representational uncertainty on VWM performance.

Main Methods:

• Developed a hierarchical computational model of VWM, extending the distance-dependent RNN concept.
• Incorporated a top-down feedback mechanism for dynamic refinement of object and feature representations.
• Utilized stability analysis (energy functions) and error rate analysis in change detection tasks to evaluate model performance.

Main Results:

• The model demonstrates robustness through stability analysis under varying input conditions.
• Performance in change detection tasks is influenced by receptive field sizes and input configurations (crowded vs. spaced).
• A novel cross-talk function modulates feature specificity based on feedback depth, reflecting representational uncertainty.

Conclusions:

• The proposed hierarchical VWM model effectively simulates object and feature processing with a prioritized object-level analysis.
• Top-down feedback mechanisms are crucial for refining representations and improving accuracy in VWM tasks.
• The model provides a framework for understanding how the visual processing hierarchy supports VWM.