Barndorff-Nielsen diffusions model neurons in primary sensory cortex

Toby Berger, Mustafa Sungkar
2016 2016 Information Theory and Applications Workshop (ITA)  
Neural information is Shannon information! We believe neurons in primary sensory cortex maximize the information transmitted per energy expended. Let "j" be such a neuron. Neuron j receives spike trains from circa 10,000 other neurons. In response, j produces a voltage called the post-synaptic potential (PSP). When the PSP reaches a threshold, j sends an action potential (AP) to all its target neurons, which also number circa 10,000. PSP's are modeled as time and amplitude-continuous random
more » ... ntinuous random processes of the sort usually called diffusions. Barndorff-Nielsen (B-N) diffusions are time-homogeneous, amplitudeinhomogeneous Markov processes, certain of which exhibit attraction to the threshold as the diffusion approaches it. That is the very behavior that PSP buildup exhibits. The threshold hitting times of B-N diffusions have generalized inverse Gaussian (GIG) distributions. Much research favors GIG hitting times for fixed threshold neurons. We have optimized the information-energy tradeoff of such neurons. Bits per Joule (bpJ) is also bps per Jps, i.e., capacity per watt. By finding capacity v. power curves for primary cortical neuron channels, we have linked Shannon theory and neuroscience more closely.
doi:10.1109/ita.2016.7888196 dblp:conf/ita/BergerS16 fatcat:sc4xsifndfgv5hsxjofgknnghq