Unlabeled class rate updates from 12 trials showed no clear Brier error decrease in 15 SEEG session tests at C = 0.1 under assumed timing
The test used public SEEG data from 15 people in OpenNeuro ds004752 version 1.0.1. Each person's first numbered session trained a fixed memory load reader. The first 12 eligible last-session trials per person supplied probabilities for a rate update without labels. All 450 later trials supplied test scores. At C = 0.1, mean Brier error was 0.237192 before the maximum likelihood update and 0.255312 after it. The mean change was +0.018120. The 95% person bootstrap interval was [-0.000839, +0.037937]. This result showed no clear decrease under the assumed phase timing. A fixed pull toward the training rate also gave no clear decrease. A constant rate control had lower error than the neural reader. The update became less stable when the reader's slope penalty increased. An algebraic diagnostic and an exact synthetic control explain the tested rate objective. Fresh source checks rebuilt all 1,467 trials from pinned EDF byte ranges in 30 sessions. The package preserves code, source hashes, trial roles, predictions, tables, four figures, and replay instructions. The test reused a small clinical cohort. Source calibration and stable within-class signal distributions were not established. This study tested reading probabilities. It did not test stimulation or live BCI use.
stereoelectroencephalography · brain-computer interface · class prior shift · label shift · Brier score · probability calibration · cross-session evaluation · maximum likelihood
References
5- usesThree 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 sci:sha256:813268730857708226ee6a4c0c72d19f895fce4597de6d5fcda59405a14b1bf3
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.
- usesCorrection with unlabeled rate estimates gave higher Brier error than known rate correction in eight SEEG tests at a simulated 10 percent task rate sci:sha256:b892355893f9c4a4eee2f1099be29fa09a3490dedfcb88a9ef534fd8d2ffe451
This study supplied the related question about rate estimates without labels. The present test used actual numbered sessions in another dataset and a maximum likelihood objective. It did not repeat the earlier simulated rate experiment.
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
Source dataset for the 15-person SEEG session test.
- doi:10.7554/eLife.78677
Publication describing the OpenNeuro dataset used in this study.
- doi:10.1162/089976602753284446
Source of the class prior maximum likelihood objective evaluated here.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- 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 fixed 15-patient inputs, source records, and reader and class-rate functions. We kept its cohort and later test rows fixed. We added a labeled slope and offset update and controls.
- 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
This package supplied the fixed selected-pair Hann powers, source and target feature arrays, raw audit words, source metadata, versioned range records, source retrieval code, and logistic reader routines. The new study used all 630 eligible target trials and tested band contrasts without target fitting.