%0 Journal Article
%T Universality of Power Law Coding for Principal Neurons
%A Gabriele Scheler
%J Computer Science
%D 2014
%I arXiv
%X In this paper we document distributions for spike rates, synaptic weights and neural gains for principal neurons in various tissues and under different behavioral conditions. We find a remarkable consistency of a power-law, specifically lognormal, distribution across observations from auditory or visual cortex as well as midbrain nuclei, cerebellar Purkinje cells and striatal medium spiny neurons. An exception is documented for fast-spiking interneurons, as non-coding neurons, which seem to follow a normal distribution. The difference between strongly recurrent and transfer connectivity (cortex vs. striatum and cerebellum), or the level of activation (low in cortex, high in Purkinje cells and midbrain nuclei) seems to be irrelevant for these distributions. This has certain implications on neural coding. In particular, logarithmic scale distribution of neuronal output appears as a structural phenomenon that is always present in coding neurons. We also report data for a lognormal distribution of synaptic strengths in cortex, cerebellum and hippocampus and for intrinsic excitability in striatum, cortex and cerebellum. We present a neural model for gain, weights and spike rates, specifically matching the width of distributions. We discuss the data from the perspective of a hierarchical coding scheme with few sparse or top-level features and many additional distributed low-level features. Logarithmic-scale coding may solve an access problem by combining a local modular structure with few high frequency contact points. Computational models may need to incorporate these observations as primary constraints. More data are needed to consolidate the observations.
%U http://arxiv.org/abs/1410.5610v1