Training on six other people reduced unseen contribution error by 4.94% in 250 SEEG records. The centered target used a 7 Hz, 1 s filter at a 50 ms delay.
We tested prediction beyond exact known filter input in 250 public hippocampal SEEG records from seven people with epilepsy. The target used a centered 7 Hz filter with 1 s support and a 50 ms center delay. Its offline label required voltage through 550 ms after the cutoff. Each linear reader used early training data from six other people and excluded every record from its test person. The test record supplied its own early voltage mean and scale. Mean relative squared error reduction for the unseen contribution was 4.94%. All seven person means were positive. The descriptive 95% person resampling range was 2.91% to 6.31%, with donor fits repeated. Mean phase error fell from 54.35 to 52.65 degrees. Independent Gaussian controls gave slightly negative skill. These reused device-event recordings do not establish neural phase, a stimulation effect, or prospective transfer. The package preserves native data, source pins, fitted weights, all test forecasts, four figures, code, and independent checks.
stereoelectroencephalography · signal prediction · filter target · cross-person model · phase error · epilepsy
References
6- usesZero sample runs continued after all 932 marked stimulation intervals in 250 selected hippocampal SEEG records sci:sha256:08b79fb28c0ad8ae5ce5f97718a82ab9495d0a888ebc9474e04e519df67d70f3
This work supplied the inherited record timing rules and source selection ledger. Those rules permit the early fits and later reference targets to end before marked stimulation.
- usesA fixed 50 ms phase forecast had 73.87 degrees more error with a centered target in 250 SEEG records at 7 Hz with a 1 s Hann filter sci:sha256:705b9d05983c9f566daeb08f62e6e1d54faaa1a424497602d6d50cff1eabb101
The restored parent supplied the 250-record cohort, exact centered-target convention, native data, source receipts, EDF reader, and retrieval code. This study adds prediction of the unseen contribution using six other people.
- usesA 7 Hz causal filter reader reduced unseen input squared error by a mean 50.4 percent at 50 ms in nine SEEG rest records sci:sha256:f851bd1de5c8367288ad9745a4316baf6f4e7fde8dfec5bb56d0adb133df2ac4
This work supplied the benchmark concept of comparing a predicted unseen contribution with the exact known filter contribution. We use this score concept for a centered finite filter and exclude each test person from its donor fit.
- usesAn exact check gave more 7 Hz phase certificates after SEEG gaps in 250 records with bounded missing voltage sci:sha256:30f5c0e1c4269dd470b69e543ec39d5cf9cefcb91eac8e1b9a828f4dc9cbe095
This source package supplied the selected native SEEG records, source version receipts, EDF reader, and retrieval code inherited through the restored parent. We kept all 250 selected records and made no new voltage selection.
- usesdoi:10.18112/openneuro.ds007095.v1.0.0
Analyzed source data are OpenNeuro ds007095 version 1.0.0, RNS_Epilepsy-iBIDS; the saved dataset description records CC0.
- doi:10.3389/fnins.2026.1815732
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.