Feedback with assumed correct labels at recall end lowered mean log loss and AUC in 16 reused SEEG session pairs at C=0.01
We tested delayed recall feedback for a fixed SEEG memory reader. The test used 16 previously used session pairs from OpenNeuro ds004809 version 2.2.0. We assumed that correct labels were available when each recall period ended. The first five target lists supplied 60 labels for calibration. Later updates used only labels from completed lists. At C=0.01, mean log loss on 2,880 test words fell from 0.606145 to 0.596587. The mean gain was 0.009558. The 95% person bootstrap interval was 0.001201 to 0.019449. Nine of 16 people had a positive gain. Mean AUC across test words fell from 0.595693 to 0.584912. Word order within each list stayed the same. A task-only reader had a similar loss gain. The test did not establish an added gain from SEEG. The package contains code, features, source records, retrieval instructions, checks, tables, and figures. This retrospective test did not apply brain stimulation.
delayed feedback · memory recall · SEEG · probability calibration · log loss · retrospective analysis
References
4- usesAn added SEEG model gave an uncertain loss gain in 16 later sessions at C=0.01 sci:sha256:55fcb7452d9209238a81470c6f0c0fa4b8290482a4659458eb194cc7e0caeeb5
Used its benchmark selection, 96 power features, task inputs, source-fit objective, and pinned raw retrieval recipe through the direct parent.
- usesOne fitted neural weight per session reduced mean log loss against a fixed weight in 16 reused SEEG session pairs at C=0.01 sci:sha256:196e275c28c32d7bb9483b9d51b2fca203b4bde7689cf921f7c3b7496ee07d44
Reused all 16 session pairs, source model coefficients, calibrated neural-weight method, source records, and feature inputs. Tested a new list-delayed offset policy.
- usesCorrection from up to 20 past trials increased mean SEEG recall forecast error in 13 patients under assumed source trial order sci:sha256:53c5da5570fd39067a0e0ac07cd483616dcdd9730e7838f130b9ee564981a4e9
Used its recent-feedback offset result to define the background question. The present test uses a different dataset and checked list recall times. No data or fitted parameters were copied from it.
- usesdoi:10.18112/openneuro.ds004809.v2.2.0
The study analyzes OpenNeuro ds004809 version 2.2.0 data; the saved version-specific metadata identifies this DOI and the CC0 license.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesShared SEEG weights lowered mean recall forecast loss by 0.00173 nats/word in 16 reused session pairs after five-list calibration at C=0.1
Used its 32 fixed feature arrays, trial tables, pinned source records, and MIT task and retrieval helpers. Checked copied file hashes. Fit new shared and local six-band models; did not use its fitted neural models.