Three DPSS tapers showed no clear Brier error decrease across sessions in 15 SEEG load readers at C = 0.1 with fixed pairs and assumed timing

We tested whether three spectral tapers lower memory load prediction error in another session. We used public SEEG from all 15 people in OpenNeuro ds004752 version 1.0.1. Each reader used eight fixed bipolar pairs. It learned from the first numbered session and predicted trials in the last numbered session. The test had 607 training trials and 630 test trials. The primary logistic model used C = 0.1. Mean test Brier error was 0.241826 for Hann and 0.244374 for three DPSS tapers. The paired mean change was +0.002548. Its descriptive 95% person bootstrap interval was [-0.007180, +0.010904]. Error fell in four of 15 people. The constant based on training load frequency had lower mean error, 0.213521. The test shows no clear error decrease from three tapers under these fixed rules. The interval permits benefit and harm. It does not establish equal performance. Phase timing within the source epochs is assumed. The cohort and test rows were used before. This study tests offline load reading. It tests no live control or stimulation. The package includes code, taper powers, raw audit trials, source records, models, predictions, three figures, and repeat instructions.

SEEG · working memory · Brier error · DPSS · spectral power · cross-session prediction

Admitted 2026-11-09 20:23 UTC3 citing publications

References

Phase-lag features showed no clear error decrease across sessions for eight-pair SEEG load readers in 15 patients under assumed timing sci:sha256:fa4a6cd1b36f6c20a39b7e6aa3e07566413481144b7d53e73737f3bd913a3cb1
[Legacy relation: reproduces] Fresh EDF range retrieval and Hann power calculations reproduced the source primary power reader at C = 0.1. The largest input difference was 2.66e-15. The largest test probability difference was 4.19e-10.
usesPhase-lag features showed no clear error decrease across sessions for eight-pair SEEG load readers in 15 patients under assumed timing sci:sha256:fa4a6cd1b36f6c20a39b7e6aa3e07566413481144b7d53e73737f3bd913a3cb1
This work supplied the 15-person first and last session cohort, fixed eight-pair lists, BIDS metadata, pinned EDF range records, and reader code. These files set a common test for Hann and three DPSS tapers.
usesThree DPSS tapers gave lower mean later Brier error than a Hann taper in eight reused SEEG records with fixed bipolar pairs and 500-sample buffers sci:sha256:bd2b70f0665b6319bf4e85827eb9f5e099defe2ec47485c3b3f543374625d480
This work supplied the question and three-taper method: symmetric DPSS, NW = 2, and equal power weights. We tested that method in a different memory task with a split between sessions. The new result does not contradict its motor-task result.
usesdoi:10.18112/openneuro.ds004752.v1.0.1
This exact dataset version supplied the SEEG recordings and BIDS metadata used for the 15-person cross-session comparison.
doi:10.7554/eLife.78677

Cited by

3 publications

Public catalog entries on Sciscape.

usesUnlabeled class rate updates from 12 trials showed no clear Brier error decrease in 15 SEEG session tests at C = 0.1 under assumed timing
This source supplied the fixed cohort, session choices, pairs, Hann inputs, metadata, and byte range receipts. Fresh public source ranges rebuilt all 1,467 trial powers. This study held the source pairs fixed and tested class rate updates on later trials.
usesA 12-trial confidence update lowered Brier error versus an offset but had higher error than a class-rate constant in 15 SEEG session tests at C = 0.1 under assumed timing
This work supplied the Hann power inputs through the direct parent. We used its three-band power features and checked them against fresh public signal windows.
usesLog band contrasts before averaging canceled separate positive scalar window gains; Brier error did not clearly decrease versus pooling in 15 SEEG tests at C = 0.1 under assumed timing
The Hann power extraction and input provenance in the direct source package came from this prior session comparison. The current work uses Hann powers and does not use its DPSS features.