Joint 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
We tested whether the relation between two uncertain scores improves command probabilities after known simulated contact loss. We used public SEEG data from OpenNeuro ds004473, version 1.0.2. Early trials supplied all fitted values. We compared joint Gaussian score integration with multiplication of the same two marginal probability estimates. The test used 674 later trials from eight people and 111 physical contact faults. Mean joint log loss was 0.91955 with joint integration and 0.91952 with marginal multiplication. The mean increase was 0.00003 nats per window. The exact two-sided sign-flip value was 0.117. This small change does not show a practical BCI benefit. An exact probability identity and a model example show why marginal multiplication can change a command. The package preserves an observed-input reader, three figures, feature data, model and fault states, predictions, source records, and reproduction code. This is an exploratory follow-up on a reused clinical cohort. It does not test stimulation or live use.
brain-computer interface · SEEG · missing inputs · Gaussian integration · contact loss · command probabilities
References
6- 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
This publication supplied the conditional feature recovery rule and fixed 0.5 diagonal covariance shrinkage. We extended it to conditional uncertainty in two command scores.
- usesGaussian score integration reduced mean log loss after known simulated shaft loss in eight SEEG recordings sci:sha256:df8af524f911208eacdc37c6b24abd4771da296cb51cf5131292114becc408c3
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.
- 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 publication first preserved the feature arrays, extraction code, trial exclusions, and pinned raw source records used in this study.
- usesTwo tuned SEEG readers gave mean log losses of 0.906 and 0.908 in eight later-trial tests sci:sha256:427e0b56a373b2720735fa431b8d69a55a3dab504252b9c81ac01a178df23f3c
This publication supplied the exact SEEG feature files, source metadata, frozen primary two-head readers, chronological split, and raw source check functions. We tested those readers under known simulated contact loss.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
Analyzed public SEEG dataset; version 1.0.2 is recorded in the source package.
- doi:10.1088/1741-2552%2Facae0a
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesCovariance from remaining inputs increased mean SEEG accuracy for eight sites from 27.1% to 34.7% after known simulated shaft loss in three offline sessions at 100 to 500 ms
Its known-input-loss comparison informed the question about Gaussian treatment of missing features. We developed a direct common-covariance site-score comparison for ds004457. We used no trial data, code, or numerical results from that study.
- usesCovariance from remaining inputs increased mean SEEG accuracy for eight sites from 27.1% to 34.7% after known simulated shaft loss in three offline sessions at 100 to 500 ms
Its known-input-loss comparison informed the question about Gaussian treatment of missing features. We developed a direct common-covariance site-score comparison for ds004457. We used no trial data, code, or numerical results from that study.