Bits assigned from fixed reader weights reduced mean coding error by 18% at 32 bits per update on 100 later passages in nine SEEG listeners
This study tested feature coding for fixed SEEG sound readers. We fitted linear readers with eight pairs from six early passages per person. We fixed each stream's limits and bit widths before testing 100 later passages in nine people. At 32 bits per update, bits assigned from reader weights reduced mean coding MSE from 0.00102044 to 0.00083815. This was a 17.9% reduction. Coding MSE measures change from the same fixed reader with float features. It fell in seven people, rose in one, and stayed equal in one. The descriptive person bootstrap interval for the mean change was [-0.00030943, -0.00007031]. At 10 updates per second, the feature payload was 320 bits per second. This was 87.5% less than eight float32 streams. The test did not show better sound prediction. Mean target MSE change was +0.00014643; its interval included zero. Allocation had higher mean coding MSE at 16 bits. Fixed limits also clipped later features. The archived package contains an exact bound allocator, real bit packets, figures, source-derived inputs, and source retrieval code. This SCI package retains the allocator, packets, figures, and source code, but omits the recordings, audio, event tables, and source-derived input caches because of conflicting dataset terms. All used feature and target arrays matched a fresh reconstruction from pinned public source ranges. This is an exploratory reuse of public ds004703 v1.1.0. It does not test intended speech, hardware energy, or stimulation.
SEEG · feature coding · bit allocation · quantization · brain-computer interfaces · speech amplitude · ridge regression
References
6- usesEarly score costs reduced mean probability change for eight fixed SEEG readers at a 32-bit sample budget sci:sha256:85475ddee4e5c22adeb1635ed39e98827abfe25c5349b132743b845078a43b0f
Its fixed-reader bit coding test motivated the question. The present test uses continuous sound outputs, fixed early limits, and an analytic bound. No code or data was copied from that publication.
- usesFixed SEEG readers had higher mean sound amplitude correlation at 400 ms than at zero delay on 100 unused later passage records in nine people sci:sha256:3c8b962f91454c5d1b32469fb10a6ecb9534841b65f6b7ab946bc56a062e2126
This study used its 154 corrected feature records, source receipts, clock offsets, later-passage rule, and retrieval code. New readers used one power band.
- usesLocal power ratios gave 52.13 percent mean balanced accuracy for heard sound classes with eight SEEG pairs in later excerpts from ten people sci:sha256:3dd9a550bef98a8cb24fc94fac327140049f0aa8d6f97960740b338ae8b8f777
The inherited EDF parser and adjacent-contact pair rules were adapted from this publication. The package preserves their code and MIT notices.
- usesTen offline SEEG readers gave mean correlations of 0.214 at zero added delay and 0.302 at 400 ms for heard sound amplitude sci:sha256:0363ee7f6d1912a2c732d3938bfa9b204378683c1ded14e573a003ec7c3ef2fc
This package retains its EDF and WAV retrieval code and causal feature rules through the corrected-input publication. The MIT notice is preserved.
- usesdoi:10.18112/openneuro.ds004703.v1.1.0
The recorded source dataset and version used for the passive listening recordings and derived inputs.
- doi:10.1038/s41467-024-44844-9
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesGuarded difference coding increased mean squared reader error by 79% at 32 bits per update on 100 later passages in nine SEEG listeners with correct packet delivery
Used the complete archived input package, nine-person split, fixed reader rules, and assigned-bit comparator. All 154 feature files were rebuilt from pinned source ranges. The new test compares guarded difference coding against the absolute and assigned-bit methods.