Four-bit storage reduced waveform payload by 87% and increased mean normalized 20–500 ms forecast error by 0.70% compared with float32 in eight SEEG sessions
We tested four-bit storage of pulse-response forecasts in public OpenNeuro ds004696 version 1.0.1. Five early pulses fitted each forecast. Five later pulses tested it. The test used 886 recording pairs at 86 stimulation sites in eight patient sessions. Four-bit storage increased mean normalized single-pulse error by 0.701% relative to float32 storage. The descriptive 95% patient bootstrap interval was 0.406% to 1.072%. Waveform payload fell by 87.39%. The observed mean was below the fixed 1% margin. The one-sided 95% Student t upper bound was 1.052%. Thus the test did not establish that the mean was below the margin under independent normal patient-effect assumptions. Mean raw voltage error increased by 4.15%. Some recording pairs had large error increases. The package preserves real packed models, exact test products, source hashes, code, and figures. This is a fixed exploratory test on reused data. It does not test live control, a new day, or behavioral effects.
stereoelectroencephalography · brain-computer interface · pulse-response forecast · four-bit quantization · waveform storage · OpenNeuro
References
5- usesFive-pulse noise scaling reduced later single-pulse error for 20–500 ms centered bipolar SEEG in eight patient sessions sci:sha256:fbc06ab469d48eac7a6f54dd97cd24ccd9c7c5ca4b0db4d64340536b3e81428a
This publication supplied the five-pulse noise-scale model, eligible pair identities, exact trial products, and source provenance. We rebuilt all eight source files and tested amplitude rounding of its forecasts.
- usesA five-pulse check selected more repeatable bipolar SEEG pairs within eight epilepsy patient sessions sci:sha256:b5b7963746f5c6d2a2a8f7844ec590732a4eb152bc7a70979b3bc95c3e27820f
The original extraction recipe defined the adjacent bipolar pairs, good-pulse rules, source row mapping, and exact time windows. We preserved that recipe through the noise-scale publication.
- usesTwenty cosine coefficients increased mean 20–500 ms SEEG forecast error versus full scaled templates in eight patient sessions with five calibration pulses sci:sha256:f0178c1dc521dcf46916de948b41e3ce68a95c5741364315354453d40f62b230
The cosine-compression study supplied the motivation and the prior result read before this study. We tested fixed amplitude codes on the same source trials. This is not an independent cohort or a direct new comparison of the two methods.
- usesdoi:10.18112/openneuro.ds004696.v1.0.1
The study used author-supplied MATLAB derivatives from OpenNeuro ds004696 version 1.0.1.
- doi:10.1523/JNEUROSCI.2201-22.2023
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesA fixed joint SEEG model increased mean normalized error compared with separate scaling after five calibration pulses at 20–500 ms in eight sessions
This archive supplied the stimulation site and readout selection tables, exact within-readout pulse products, source metadata, and the source retrieval code. We rebuilt every inherited pulse product from the public source and added exact cross-readout products for a fixed spatial forecast test.
- usesPacked SEEG signs reduced float32 waveform bytes by 96.9 percent and mean response AUC by 0.028 in eight sessions at 20 to 500 ms
This earlier forecast study informed the scope distinction between model memory and binary waveform input.