Inferring the capacity of the vector Poisson channel with a Bernoulli model

Area of Science:

• Computational neuroscience
• Information theory
• Neural coding

Background:

• The capacity of information transmission channels sets fundamental limits on signal fidelity.
• Understanding neural population structures is crucial for decoding neural activity and its information-carrying capacity.

Purpose of the Study:

• To derive the information transmission capacity of parallel Poisson process channels.
• To evaluate the relative effectiveness of different neural population structures for information processing.
• To investigate the impact of neural coding strategies on stimulus reconstruction fidelity.

Main Methods:

• Derivation of channel capacity for parallel Poisson process models, using Bernoulli process surrogates.
• Application of Shannon's rate-distortion theory to analyze stimulus reconstruction error.
• Modeling multi-neuron recordings as a sum of neural populations to assess capacity reduction.

Main Results:

• Inter-neuron dependencies increase capacity in neural populations with individual innervation but decrease it when inputs are shared.
• Mean-squared error of decoded Gaussian stimuli decreases exponentially with population size and maximal discharge rate.
• Modeling multi-neuron recordings as a summed population significantly reduces capacity compared to individual neural responses.

Conclusions:

• Population coding is essential for accurate stimulus reconstruction in neural systems.
• Attempting neural control without spike sorting drastically reduces achievable information fidelity.
• Single-electrode neural stimulation maintains fidelity comparable to stimulating individual neurons.