A neural network model for visual selection and shifting.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Computer Vision

Background:

• Simulating visual selection and attention shifting is crucial for understanding cognitive processes and developing advanced AI.
• Traditional models often rely on complex differential equations and limit cycles to represent neuron oscillations, posing simulation challenges.

Purpose of the Study:

• To develop a novel two-layer neural network model for simulating visual selection and attention shifting.
• To introduce an instantaneous frequency mapping dynamic model as an alternative to limit cycle-based neuron oscillation simulation.

Main Methods:

• A two-layer neural network was constructed.
• An instantaneous frequency mapping dynamic model was developed to describe neuron frequency changes.
• The first layer implemented image segmentation, and the second layer acted as a visual selector.

Main Results:

• The model successfully simulates the mechanism of visual selection and shifting.
• Neuron activity was effectively rebuilt based on instantaneous frequency.
• The dynamic interaction between the central neuron and object-representing neurons facilitated object selection and attention shifting.

Conclusions:

• The instantaneous frequency mapping dynamic model provides a viable alternative for simulating neuron activity in visual selection tasks.
• The proposed two-layer network effectively models the dynamic process of visual attention shifting and object selection.