Nonlinear Hebbian Learning as a Unifying Principle in Receptive Field Formation

Area of Science:

• Computational Neuroscience
• Neuroscience
• Machine Learning

Background:

• Sensory receptive field development is explained by diverse models like sparse coding and synaptic plasticity.
• Existing models lack a unifying principle for receptive field formation.

Purpose of the Study:

• To unify diverse models of receptive field development under a single principle.
• To investigate the role of input statistics and nonlinearities in receptive field formation.

Main Methods:

• Application of nonlinear Hebbian learning to natural image data.
• Analysis of receptive field development across different neuron models and plasticity rules.
• Extension of the model to auditory and V2 visual processing.

Main Results:

• Nonlinear Hebbian learning unifies normative and bottom-up models of receptive field development.
• Receptive field shapes are constrained by input statistics and preprocessing, with minimal variation across nonlinearities.
• Localized receptive fields develop without requiring overcompleteness or sparse network activity.

Conclusions:

• Nonlinear Hebbian learning provides a universal framework for understanding receptive field formation.
• Natural statistics and nonlinear Hebbian learning explain key aspects of receptive field development across sensory modalities.