Gaussian score integration reduced mean log loss after known simulated shaft loss in eight SEEG recordings
We tested task probabilities after a known simulated shaft loss in public SEEG data. Eight people supplied 2014 retained trials. Each reader used early trials for fitting and 674 later trials for testing. Gaussian score integration reduced mean log loss from 0.58438 to 0.57769. Seven people had lower loss. The mean decrease was 0.00668, with a descriptive person-bootstrap interval from 0.00251 to 0.01178. Decisions at 0.5 and AUC did not change. A fixed common variance per person gave lower mean loss, 0.57356, than a separate variance for each lost shaft. Thus, the decrease supports probability smoothing but does not validate detailed fault uncertainty. The package preserves code, copied features, fitted readers, fault probabilities, source versions and hashes, audit bytes, tables, figures, and retrieval instructions. Independent source and numerical checks passed. This exploratory test reused a clinical cohort and used known event times, future-informed trial screening, and unchanged surviving inputs. It does not establish live BCI or stimulation performance.
stereoelectroencephalography · signal loss · probability calibration · logistic regression · Gaussian integration · brain-computer interface
References
5- usesTraining covariance increased mean balanced accuracy by 0.36 percentage points after known simulated loss of one contact in eight offline SEEG decoders sci:sha256:def7ff6139316b5d2345596c08c24a0bc6d86abe4fa6a51d420c77ccb5674e3d
Its covariance conditional-mean reader supplied the comparison method and fixed shrinkage rule. We added probability integration and shaft-loss tests.
- usesWhole-shaft training showed no clear gain over matched pair omission in lowest balanced accuracy after known shaft loss with mean replacement on later trials from eight SEEG recordings sci:sha256:89f21dfe80a55b2dc64510c3cd6a5fe87920820adc06f046950a27d862988c48
Its whole-shaft setup informed the known shaft masks. We tested later probability loss rather than shaft omission during fitting.
- usesA rest reference based on the mean of three power bands gave 3.21 percentage points higher mean later SEEG balanced accuracy than full subtraction of one rest window in eight recordings sci:sha256:af9e496839fdbe03c592f73a9ec02b28dbec47ca25a37e3d44e1a29184a2acec
We used its exact feature arrays, source metadata, audit ranges, and retrieval code. These files supplied the public SEEG inputs and fixed source provenance.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analyzed SEEG features and source records trace to OpenNeuro ds004473 version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesScore integration gave lower mean Brier error than a Gaussian class reader after simulated shaft loss in eight SEEG recordings with 64 calibration trials
Supplied the Gaussian score integration comparison, conditional score formulas, common-variance control, and known shaft loss setup. The new test uses fewer calibration trials and a condition change.
- usesJoint Gaussian score integration increased mean command log loss by 0.00003 nats per window after one known simulated contact loss in eight offline SEEG readers
This publication tested Gaussian probability integration for a binary SEEG activity reader. We extended the question to joint activity and key probabilities and isolated cross-score covariance.
- usesWorst-mask SEEG training reduced mean worst-shaft log loss at C = 1 in eight tests with 32 fit trials
We use the known single-shaft masks and survivor-bank comparison design. We change the training split and test a single reader trained on maximum scenario loss.
- usesWorst-loss training gave mixed later error changes compared with mean-loss training under known single-shaft loss in eight SEEG recordings
The known simulated shaft-loss setting provides the fault context. This study compares common fixed training objectives instead of correction of a score after loss.