A 64-window rolling rank map raised mean Brier error in eight later SEEG tests
We tested a rank map that updates without task labels. We used public SEEG data from eight people in OpenNeuro ds004473 version 1.0.2. All readers used the same eight signal pairs and fitted weights. The main test used 674 later trials. Mean class-balanced Brier error was 0.1891 with a fixed map and 0.1916 with a map based on the last 64 windows. The mean increase was 0.002419. The descriptive 95 percent person bootstrap interval was [0.000563, 0.004585]. A person sign-flip reference gave p = 0.0547. The rolling map reduced error under an imposed gain ramp. Its error increased when replay sampling changed the state frequency, with no imposed gain. An exact two-value example shows why a change in state frequency can shift rank values. The package includes the features, fitted models, window probabilities, three figures, source code, and public retrieval instructions. A raw-range rebuild matched all 3,133,395 feature values. These results concern this reader and these sampled within-session states. They do not establish a population effect or a live BCI result.
stereoelectroencephalography · brain-computer interface · rolling rank map · Brier error · state frequency · gain change · controlled replay · intracranial EEG
References
3- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
We used all eight feature sets, trial records, source range hashes, and the extraction code. A fresh public raw-range rebuild matched every saved feature value. These data support the new rolling-map and replay comparisons.
- usesTraining rank maps gave no clear probability error gain in eight later-trial SEEG tests sci:sha256:58757a3770e41c3615f92d058ded90aec51044808c10260ec7ba6bbc27cae3ce
We used the fixed training-rank rule, fitting settings, pair choice, and later-trial split as the fixed reader. Our 1,348 fixed-map probabilities matched its archived rank-reader probabilities within 1.3e-13. We added causal rolling maps and controlled gain and state-frequency replays.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used OpenNeuro ds004473 version 1.0.2 for its SEEG recordings and task metadata; the publication preserves version-specific retrieval records and hashes.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.