sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A fixed Gaussian template correction reduced four-pulse site log loss in five offline SEEG sessions at 20 to 100 ms

We tested a fixed probability rule on public SEEG data from OpenNeuro ds004457 version 1.0.2. Three early pulses per site fitted each template. A fourth pulse fitted a common probability scale. The final four pulses tested 183 sites in five patient sessions. A Gaussian model of shared template error gives a factor of 4/7 for four-pulse sum evidence. This factor was fixed before the new scores were computed. Mean site log loss fell from 2.891 to 2.592 bits. The gain was 0.299 bits and was positive in all five patients. The descriptive patient interval was 0.238 to 0.383 bits. The exact two-sided patient sign-flip test gave p = 0.0625. Mean Brier error fell from 0.6254 to 0.5971. Site accuracy stayed at 48.3%. The correction gave mixed log-loss changes versus division by four. That small mean gain changed sign without the final probability mixture. These reused offline data do not identify the cause of probability error or establish online brain control. The package preserves code, exact voltage inputs, source hashes, all probabilities, two figures, and reproduction checks.

SEEG · stereoelectroencephalography · Gaussian model · template error · probability calibration · brain interfaces

Admitted 1 citing publication

References

5
  1. usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886

    This archive supplied verified voltage bins, fixed contact and site eligibility, source metadata, and the native extraction code. We rebuilt all inputs from pinned raw files and tested a new repeated-pulse probability correction.

  2. usesDivision of evidence by four reduced mean log loss with a fixed Gaussian reader in five offline SEEG sessions sci:sha256:0e0e0f6a34314ea79c4cbfc48b9b3ac8a317696a1d0e1a22699e3b9a775e05cc

    We used its three-template-pulse reader, separate fourth-pulse temperature fit, and fixed sum and mean evidence rules. We read its results before the new analysis. This extension tests the factor (m+1)/(m+n) from shared Gaussian template error. It reuses the same patients and test pulses. The preserved sum and mean baselines match within 3.7e-7 in log probability.

  3. usesMeans of two responses lowered log score per pulse for calibrated site readers in five SEEG sessions at 20 to 100 ms sci:sha256:418d518461019ac1084a4712431f8bb4e3729685bb28406d36bb68f54d131a61

    Its pulse-cost analysis motivated reporting the number and duration of responses used by an episode. Our test concerns probability scale with an unchanged site decision. The site set and fitting budget differ, so we make no direct score comparison with that report.

  4. usesdoi:10.18112/openneuro.ds004457.v1.0.2

    Source SEEG recordings analyzed in this publication; the exact version and CC0 license are recorded in the pinned dataset metadata.

  5. doi:10.1523/JNEUROSCI.1325-22.2023

    Source study describing the ventral temporal cortex recordings used in the dataset.

Cited by

1 publication

Public catalog entries on Sciscape.