Efficient coding correlates with spatial frequency tuning in a model of V1 receptive field organization

Area of Science:

• Computational Neuroscience
• Computational Vision
• Machine Learning

Background:

• Efficient coding is crucial for early visual processing.
• Previous models optimized specific aspects like mutual information or sparseness.

Purpose of the Study:

• To investigate how different efficient coding estimates impact model receptive field similarity to V1 data.
• To analyze the role of nonlinear dynamics and Hebbian learning in efficient coding models.

Main Methods:

• Simulated a model with nonlinear dynamics and Hebbian learning.
• Applied various efficient coding estimates as objective functions.
• Compared model receptive fields to experimental V1 data based on spatial tuning.

Main Results:

• Most efficient coding measures significantly correlated with increased model-data similarity.
• The degree of correlation varied across different efficient coding estimates.
• Firing pattern variance of individual cells did not predict model-data similarity.

Conclusions:

• Optimizing efficient coding generally enhances the biological plausibility of computational models.
• The choice of efficient coding metric influences the degree of similarity to neural data.
• Firing rate variability is not a reliable indicator for efficient coding in this context.