Added SEEG power increased recall forecast loss in all eight later sessions with a fixed logistic reader
We tested recall forecasts in eight people from public SEEG data in OpenNeuro ds004809, version 2.2.0. A source session trained each fixed logistic reader. The first five lists of the next numbered session calibrated its intercept. The last 15 lists supplied 180 test words per person. The task reader used word position, category, and list number. The joint reader added 96 power features from 16 bipolar pairs. At C = 0.1, mean loss increased from 0.6714 to 0.8733 nats per word. It increased in all eight people. The mean task-minus-joint gain was -0.2019. Its descriptive 95 percent bootstrap interval was [-0.2656, -0.1427]. A fixed C = 0.01 check reduced the extra mean loss to 0.0165. We preserve code, features, models, forecasts, figures, and pinned data retrieval records. An independent EDF reader checked 46,080 feature values from one complete session pair. A separate analysis run reproduced all scores and model arrays. These results apply to the fixed readers and selected sessions. They do not measure a stimulation effect or establish that SEEG has no recall information.
stereoelectroencephalography · memory recall · probability forecast · logistic regression · session transfer
References
2- usesA fixed SEEG reader had mean recall AUC of 0.547 for random words and 0.544 for complete lists in eight sessions sci:sha256:853393b59d05a011f54cf6ca9a8e345e41914f3d36a0cde98355d02f5daa1c3d
We reused the EDF byte-range reader and the six-band power feature rule. We added paired-session selection and trained new readers.
- usesdoi:10.18112/openneuro.ds004809.v2.2.0
Source recordings and task events came from OpenNeuro ds004809 version 2.2.0.
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesThe mean non-recall rate was 70.1% for a sequential SEEG and task rule with three selections per list in 16 later sessions
The direct parent retained this paired-session recall design, task controls, and source retrieval lineage. We preserve that inherited code and metadata and use the same inputs for the new quota test.
- usesAn added SEEG model gave an uncertain loss gain in 16 later sessions at C=0.01
The source reader study supplied the fixed power-feature setup and the first eight session pairs carried into the direct parent package. We used its prior transfer-loss evidence to frame the late-label mixture question.
- usesThe mean non-recall rate was 70.1% for a sequential SEEG and task rule with three selections per list in 16 later sessions
The direct parent retained this paired-session recall design, task controls, and source retrieval lineage. We preserve that inherited code and metadata and use the same inputs for the new quota test.
- usesOne calibrated SEEG score gave an uncertain recall forecast gain over task calibration in 16 later sessions
The source task coding, one-second power recipe, and first eight patient session pairs come from this earlier benchmark through the direct parent package.
- usesA local power reference lowered recall forecast loss compared with absolute power in only three of eight later SEEG sessions at C=0.1
Uses the fixed eight-person cohort, post-word feature arrays, paired source pins, and task reader. These files define the shared later-session comparison.
- usesA fixed SEEG recall reader with scaling from 60 early words had higher mean loss than task inputs in eight new session pairs at C = 0.01
We used its feature files for eight people and its code for EDF retrieval, task coding, offsets, and scores. Its prior transfer results informed C = 0.01 before the eight additional people were selected. The original eight people remain a separate secondary group.