Origin of information-limiting noise correlations

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Sensory processing

Background:

• Sensory discrimination depends on information in neural populations.
• Trial-to-trial variability, specifically noise correlations, reduces this information.
• The origins of noise correlations in the brain remain largely unknown.

Purpose of the Study:

• To analytically and computationally investigate the generation of noise correlations.
• To propose a generative model for noise correlations consistent with experimental data.
• To understand the impact of noise correlations on sensory information processing and discrimination.

Main Methods:

• Analytical derivations of noise correlation patterns.
• Simulations of neural network models, specifically in V1 (primary visual cortex).
• Comparison of model predictions with neurophysiological and behavioral data.

Main Results:

• Optimal computation on information-limited inputs naturally produces noise correlations.
• These correlations explain experimental observations and lead to information saturation in large populations.
• A generative model consistent with neurophysiology and behavior was developed, without assuming suboptimal processing or internal variability.

Conclusions:

• Noise correlations arising from optimal computation on sensory input are a key factor in neural representations.
• This framework explains sensory discrimination performance and offers predictions for distinguishing sources of neural variability.
• Sensory periphery noise significantly influences cortical representations in discrimination tasks.