Weak input correlations reverse noise variance order in selected Allen circuit models with fixed wiring summaries and spectra

We tested whether mean noise variance order in equal-weight linear circuit models survives weak input correlation. We reused 247 mouse VisP graphs and 17 human TCx graphs from Allen Synaptic Physiology r2.1. Exact reference sets fix labeled degrees, class totals, specified motifs, and signed spectra. Independent input gives variable mean variance in three mouse graphs and no human graph. Unit-marginal input with one positively correlated pair at correlation 0.01 gives variation in nine mouse graphs and one human graph. All three mouse graphs with unequal independent-input variance permit an order reversal below 0.01. In a six-cell witness, correlation 0.001717314 reverses the prior order at gain 0.15. Exact algebra gives a crossover that starts with g^3/2 at small gain. Independent covariance solves verify all reference-set ranges. The effects are small model differences. Input correlations, weights, and common decay are assumptions; no linked activity data test them. The witness has an isolated inhibitory cell. These results do not establish a change in neural information or a rule across species.

Allen Institute · input correlation · stationary covariance · linear circuit model · sensitivity · signed spectrum