Optimal population coding by noisy spiking neurons

• DOI: 10.1073/pnas.1004906107
• Publication Date: 2010-07-27T03:41:37Z
• Authors: Gašper Tkačik; Jason S. Prentice; Vijay Balasubrama...; Elad Schneidman
• Keywords: Stimulus (psychology); Decorrelation; Neural coding

In retina and in cortical slice the collective response of spiking neural populations is well described by "maximum-entropy" models which only pairs neurons interact. We asked, how should such interactions be organized to maximize amount information represented population responses? To this end, we extended linear-nonlinear-Poisson model single include pairwise interactions, yielding a stimulus-dependent, maximum-entropy model. found that as varied noise level distribution network inputs, optimal smoothly interpolated achieve functions are usually regarded discrete--stimulus decorrelation, error correction, independent encoding. These reflected trade-off between efficient consumption finite bandwidth use redundancy mitigate noise. Spontaneous activity stimulus-induced patterns, single-neuron variability overestimated Our analysis suggests rather than having coding principle hardwired their architecture, networks brain adapt function changing stimulus correlations.