Pattern Adaptation and Normalization Reweighting.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• Neural adaptation alters neural response properties, including gain and preferred orientation, in response to stimulus statistics.
• In the primary visual cortex (V1), neurons exhibit suppressed responses and orientation shifts away from an adapted orientation.

Purpose of the Study:

• To propose and validate a computational model explaining the mechanisms of visual adaptation in V1.
• To investigate how homeostatic plasticity rules can account for observed changes in neural tuning during adaptation.

Main Methods:

• Developed a computational model combining divisive normalization and a simple homeostatic learning rule.
• Simulated neural responses to oriented stimuli and compared model predictions with experimental data.
• Evaluated alternative models, including those based on response correlations, covariance, and feedforward gain control.

Main Results:

• The proposed model, based on homeostatic maintenance of response products, accurately reproduces experimental data on visual adaptation.
• Dynamic adjustment of divisive normalization weights by the learning rule explains both gain changes and orientation shifts.
• The model demonstrates that a simple learning rule can achieve complex adaptive behaviors.

Conclusions:

• Homeostatic maintenance of response products through dynamic normalization is a key mechanism underlying visual adaptation in V1.
• The findings provide a parsimonious explanation for diverse physiological observations in visual adaptation.
• The model offers a framework for understanding adaptive plasticity in sensory systems.