73
good then the singular value decay can be used as a guide to estimate how large the
reduced system must be.
To test accuracy and speed, we perform a series of simulations using random inputs
and compare the reduced system’s performance with that of the full system. A match
is said to occur when a spike, indicated by a somatic depolarization of V⅛ = 40 mV
or more from rest, in the reduced system occurs within τref = 2 ms of a spike in the
full system. Our metrics for quantifying the spike-capturing accuracy of the reduced
system are the same as those in §2.7.1. We use the coincidence factor Γ (Kistler et al.,
1997), defined as
-p _ Mmatch Mfu∣]Meduced(7^ref∕T)
(Mull + Meduced)(l ~ Mfu∏(τref∕T))∕2 ’
where Muii and Meduced are the number of spikes in the full and reduced models,
respectively, Mmatch is the number of matches, and T is the length of the simulation.
To better evaluate the effect of false positives, we use the percentage matched and
percentage mismatched measures of §2.7.1, rewritten here explicitly as
% matched = 100- 1"--ch- (3.24)
MfuIi
% mismatched = 100^red-u^-^ ^match. (3.25)
’reduced
We simulate for 1 second with 200 random step currents, each lasting 0-5 ms
and having amplitudes of 0-100 pA, applied to random locations on the fiber. To