An unsupervised learning model of neural plasticity: Orientation selectivity in goggle-reared kittens

Area of Science:

• Neuroscience
• Computational Neuroscience
• Developmental Neuroscience

Background:

• Neural selectivities in the primary visual cortex are thought to reflect natural image statistics.
• Unsupervised learning models trained on natural scenes produce simple cell-like tuning.
• Existing models fail to explain orientation tuning development before structured vision begins.

Purpose of the Study:

• To investigate the role of unsupervised learning in activity-dependent visual cortex development.
• To test if computational models can explain altered neural responses in goggle-reared kittens.
• To provide a more stringent examination of models for cortical development.

Main Methods:

• Training unsupervised learning models on natural scene statistics.
• Comparing model predictions with experimental data from goggle-reared kittens.
• Analyzing neural response properties and orientation tuning.

Main Results:

• The unsupervised learning model successfully predicted orientation tuning development.
• The model accounted for altered neural response properties observed in goggle-reared kittens.
• Stimulus-driven development significantly impacts cortical responsivity.

Conclusions:

• Unsupervised learning provides a viable framework for understanding visual cortex development.
• Computational models can accurately simulate activity-dependent neural development.
• Early visual experience critically shapes neural selectivity in the primary visual cortex.