Neuronal variability reflects probabilistic inference tuned to natural image statistics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• Neuronal activity in the sensory cortex exhibits fluctuations, often viewed as noise.
• The neural sampling theory posits that this variability reflects uncertainty in perceptual inferences.
• Evidence in primary visual cortex (V1) supports this theory, but its link to natural image statistics is unclear.

Purpose of the Study:

• To investigate whether V1 neuronal variability reflects the statistical structure of visual inputs.
• To determine if probabilistic inference, tuned to natural image statistics, explains V1 response properties.

Main Methods:

• Analysis of natural image statistics.
• Electrophysiological recordings from macaque V1.
• Modeling of probabilistic inference.

Main Results:

• Natural image statistics and probabilistic inference explain the mean-variances relationship in V1 spike counts.
• Spatial context in natural images modulates V1 activity and variability consistent with the model.
• V1 response variability is explained by a probabilistic representation tuned to naturalistic inputs.

Conclusions:

• V1 neuronal variability is not mere noise but a crucial component of probabilistic inference.
• The brain's visual system is tuned to the statistical regularities of the natural environment.
• Variability in V1 responses can be understood through the lens of Bayesian inference operating on natural images.