A fixed high gamma model gave no clear recall prediction gain on later lists in 18 public SEEG sessions
We tested recall prediction in 18 public SEEG sessions from OpenNeuro ds005059, release 1.0.6. Each fixed logistic model used early lists for fitting and later lists for testing. The test set had 930 studied word pairs. The patient mean Brier score was 0.2112 with study position and list order. It was 0.2185 when study high gamma was added. The mean gain was -0.0073. The 95% patient bootstrap interval was [-0.0213, 0.0072]. The exact two-sided sign-flip p-value was 0.34576. The test did not show a clear patient mean gain from study high gamma. This result applies to the tested split, features, and fixed penalty. It does not establish a memory mechanism or clinical benefit. The package contains trial features, source versions, source hashes, code, predictions, and reproduction instructions.
associative memory · recall prediction · stereoelectroencephalography · high gamma · Brier score · patient-level evaluation · OpenNeuro
References
2- usesdoi:10.18112/openneuro.ds005059.v1.0.6
The analyzed SEEG signals and task records came from OpenNeuro ds005059, release 1.0.6.
- doi:10.1016/j.neuroimage.2015.03.077
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesWith behavior penalty 0.1, the 95% patient bootstrap interval for Brier gain included zero after neural penalty selection in 18 SEEG sessions
We reused its fixed 18-session SEEG cohort, word recall labels, power features, source pins, and extraction code. We reproduced its behavior and fixed 0.1 Brier errors and log losses within 1.3e-10. We tested earlier-list selection of the neural penalty or behavior alone while the behavior penalty stayed at 0.1.
- usesRandom list splits gave smaller mean high gamma Brier gains than later list tests in 18 SEEG sessions at penalty 0.1
We reused the verified trial features, recall labels, source versions, and Newton model code. The new tests compare random splits with its chronological benchmark in the same 18 SEEG sessions.
- usesHigh gamma models selected on early lists gave Brier gains of 0.0123 over task models and 0.0010 over recall rates in 18 SEEG sessions
We used the verified trial features, labels, source recipe, and NumPy model solver. We kept the same 18 patients and later-list split. We added early-list penalty choice and controls based on early recall rates.
- usesItem high gamma gave no clear recall prediction gain beyond list averages in 18 public SEEG patients
We reused the 18 selected SEEG sessions, compact trial power, recall labels, source reader, and pinned source records. We added list average and item deviation features and selected penalties using training lists. The patients and test lists are reused.
- usesRecall feedback raised observed mean Brier error by 0.00082 for fixed high gamma models in 18 SEEG sessions
This study uses the 18-patient SEEG trial features, labels, source records, range retrieval recipe, and logistic model solver. It preserves the original package and checks its file hashes.
- usesA fixed model with study high gamma had higher mean later recall pattern loss within lists in 18 SEEG patients
This publication supplied complete verified power inputs for 18 SEEG patients, labels, fixed source metadata, source pins and hashes, extraction and replay code, and marginal predictions for the descriptive rank comparison. The current analysis fits new conditional recall pattern models in the same cohort.
- usesThe fixed transfer model gave no clear later recall prediction gain after early calibration in nine SEEG patients
This study reused the verified trial power arrays, source identities, and source check code. It tested shared slopes in other patients after early-list intercept calibration. The same cohort was used.