A fixed rule based on taper spread gave a mean Brier change of -0.00072 in eight reused SEEG records with buffers of 500 samples
We tested whether spread among three spectral tapers can improve a short SEEG probability estimate. We used public OpenNeuro ds004473 version 1.0.2 and eight reused records. Earlier trials fit a decoder. A later block fit two calibration maps with the same number of parameters. The final 338 trials tested the maps. Mean Brier error was 0.165482 with an affine map and 0.164760 with the fixed spread rule. The mean change was -0.000723. The descriptive 95% person interval was [-0.002069, 0.000582]. Five people improved and three worsened. The result is too uncertain to support a general gain. The package preserves signals, code, figures, all fitted states, and a source retrieval recipe. The labels use known future response times. The result tests offline reading, with no test of brain stimulation.
stereoelectroencephalography · brain-computer interface · probability calibration · multitaper spectra · jackknife score spread · Brier error
References
6- 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 study uses the exact public SEEG buffers, source metadata, spectral helpers, and base model code from this package. It adds a fixed score-spread rule and matched calibration tests on later trial blocks.
- usesOne shared gain constraint gave lower mean Brier error than eight pair gain constraints in eight reused SEEG records under a uniform gain model sci:sha256:c23a39e2554b405e4ae3bc346d7725666cbf215170a730abf032b586a85f3661
This study keeps the earlier input and base-fitting method through the direct input package. It does not claim an independent test of the cohort or its final rows.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This study keeps the trial and bipolar pair rules from this source. The direct input package preserves those rules and the resulting pair lists. No new trial or pair selection is made.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The study uses the recorded OpenNeuro ds004473 version 1.0.2 source data.
- doi:10.1088/1741-2552%2Facae0a
- supportsdoi:10.1109/PROC.1982.12433
The publication cites this work for established multitaper spectra and spectrum estimation methods.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesA C sliding Fourier method kept 1,348 tested SEEG decisions with periodic Hann windows and took less time at five update intervals on one computer
This work supplied the exact eight-recording voltage buffers, the fixed pair and trial choices, pinned raw-source range records, and MIT fit and retrieval helpers. The new study fitted periodic-Hann models and compared an incremental C Fourier tool with full FFT output.
- usesAdding an intercept to scale calibration reduced later log loss in five of eight SEEG recordings at an assumed 1% task rate
We used this study as prior method context. It already applied affine calibration in a balanced taper-spread test on the same cohort. It helped define the gap: an explicit scale-only versus scale-plus-intercept comparison at an assumed rare task rate.