One random taper mixture had higher mean Brier error than a matched taper at C=1 in eight reused SEEG records

We tested a random taper mixture for a SEEG task-state reader. Each draw used one Fourier transform per bipolar pair. We used eight reused records from OpenNeuro ds004473 version 1.0.2. At penalty value C=1, mean expected Brier error was 0.189689 for the mixture and 0.185503 for a matched single taper. The primary difference was +0.004187. Its 95% person bootstrap interval was [-0.003841, +0.011776]. This interval includes both benefit and harm. The mean power across four mixture modes exactly matched the mean power from three DPSS tapers. This equality did not give lower mean reader error. At C=1, two randomly selected DPSS tapers had mean error 0.177150, compared with 0.177218 for three tapers. This secondary result does not establish equivalence. The package contains four figures, exact voltage buffers, source retrieval recipes, fitted readers, all mode predictions, and independent checks. The study concerns offline task states in reused test periods. It does not test continuous control or stimulation.

stereoelectroencephalography · spectral estimation · random taper · DPSS · brain-computer interface · Brier error