An adaptive neural model for mapping invariant target position

Area of Science:

• Computational Neuroscience
• Computer Vision
• Neuroscience

Background:

• Perception of stable objects relies on complex neural processing to overcome constantly changing sensory input.
• Egocentric target measures, based on sensory data, typically fluctuate even for stationary targets.

Purpose of the Study:

• To derive and simulate a neural network model for visual spot targets invariant to egocentric measures.
• To investigate the model's ability to learn precise sensory-motor calibrations and adapt to system changes.

Main Methods:

• Development of a neural network model representing space via movement signals.
• Simulation of the model to assess its learning, adaptability, and performance under various conditions.
• Analysis of model performance in terms of target orientation error and fault tolerance.

Main Results:

• The model achieves invariance to egocentric target measures.
• It demonstrates adaptive sensory-motor calibration, handling physical and internal parameter changes.
• Simulations show average target orientation errors of approximately 1% of the visual field, with noise and fault tolerance.

Conclusions:

• The neural network model effectively represents invariant visual targets and learns precise sensory-motor calibrations.
• Model performance suggests a strong link to the posterior parietal cortex functions.
• Testable predictions are proposed for the posterior parietal cortex's columnar topography and learning mechanisms.