Equal-amplitude models could not halve noise variance in 10 of 23 selected Allen mouse E/I input pairs

We tested noise cancellation in equal-amplitude models of directly linked Allen cortical E/I inputs. Of 23 selected mouse input pairs, 10 could not halve variance even with perfect positive correlation between magnitude fluctuations. Thirteen could do so at some correlation up to one. The equal-target, experiment and inferred-donor weighted feasible fraction was 0.595, with nominal 95% bootstrap interval 0.381 to 0.786. A shared-target versus other-target variability-matching contrast was +0.131 in 12 mouse experiments, with interval -0.168 to +0.457. Thus the sample did not resolve better matching at shared targets. Requiring 50 responses per connection left one mouse pair. One human pair cannot test transfer. The model assigns medians as means, normalizes E/I response magnitudes, and assumes linear synchronous subtraction. Covariance was not measured. This is a conditional variance bound, not a measurement of neuronal reliability. Original code, aggregate evidence, a figure, source hashes and retrieval instructions support reproduction. External records are not bundled.

Allen Institute · excitation and inhibition · noise cancellation · synaptic convergence · conditional model · response variability