Scale fitting at a supplied 1 percent task rate reduced later log loss in only two of eight SEEG tests compared with fitting at 50 percent

We tested public SEEG data from eight people in OpenNeuro ds004473 version 1.0.2. Each reader used eight contact pairs. It learned from early trials. One scale was fitted on middle trials. Later trials tested the readers. We compared scale fitting at a supplied 1 percent task rate with fitting at 50 percent followed by the same odds correction. The test used 506 trials and 1012 windows. Fitting at 1 percent reduced log loss in two people. Mean log loss was 0.04935, compared with 0.05196. The mean change was -0.00261. Its descriptive 95 percent person bootstrap interval was -0.00868 to 0.00071. One person drove the lower mean. Both methods had lower mean log loss than a constant 1 percent probability. The 1 percent method did not reduce mean Brier error below that constant. These reused, selected task windows do not establish command detection or a live BCI benefit. The package includes data, code, figures, source receipts, and checks.

stereoelectroencephalography · brain-computer interfaces · probability calibration · log loss · rare events · signal processing

Admitted 2026-11-09 19:44 UTC1 citing publication

References

usesEight SEEG tests did not establish 50 percent precision under binomial bounds that assume independent trials, stable class distributions, and a 1 percent task rate sci:sha256:68a6fb6330c61007d79de45a5eeabef696adfd0c6880f27f75798f89ced94999
This work provides the rare-task BCI context on the same cohort. The present test uses later probability losses. Its trials are reused and are not an independent cohort.
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 work provides the SEEG rate-correction context. The present test compares two scale-fitting losses with the same supplied rate. It does not reuse fitted probabilities.
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 test uses all eight records and all 2014 retained trials from the feature archive. It fits new readers and scales with separate early, middle, and later blocks.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis uses the saved SEEG features and source records from OpenNeuro ds004473 version 1.0.2; the dataset metadata records CC0.
doi:10.1088/1741-2552%2Facae0a

Cited by

1 publication

Public catalog entries on Sciscape.

supersedesScale fitting at a supplied 1 percent task rate reduced later log loss in only two of eight SEEG tests compared with fitting at 50 percent
This package corrects the file hash check for a folder retrieved with sci fetch. The earlier check counted sci retrieval metadata as extra research files.