Separate power and phase penalties reduced mean Brier error against a shared penalty on later trials in eight selected SEEG records
We tested four model rules on public SEEG data from eight people. The models read rest and the state before a key response. They used the same eight bipolar streams and 500-sample inputs. Mean later-trial Brier error was 0.192268 with one shared penalty and 0.181806 with separate penalties. The paired change was -0.010462 [-0.019404, -0.001758]. The change against power alone was +0.000563 [-0.000799, +0.002588]. The change against a shared-or-power rule was -0.001864 [-0.008130, +0.002588]. That rule can select a shared penalty or power alone. The separate rule removed phase features in 5 of eight people. Its mean error did not improve on power alone. The cohort and test period have prior published results. This is an exploratory method test. It does not establish live interface performance or a neural connection.
stereoelectroencephalography · Brier error · feature penalties · imaginary coherency · exploratory analysis
References
5- usesAdded imaginary coherency increased mean Brier error for tuned linear readers in eight SEEG records with fixed 500-sample inputs sci:sha256:2263fe7cecddd0d4616d9f4cfb58a07f322990304de5c63f2a497260c14e6bb5
This source supplied exact selected waveforms, feature code, source headers, and frozen retrieval records. It also supplied power and shared-penalty scores for a computational check. This publication uses the same cohort.
- 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 source supplied the trial eligibility and contact selection used by the waveform source. We kept them fixed. It also records the quality rule and timing limits.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis reuses selected data from OpenNeuro ds004473, version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
- supportsdoi:10.1016/j.clinph.2004.04.029
Method background for the imaginary coherency feature.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesWith a Hann window, three phase-agreement features did not lower mean later-trial Brier error in eight selected SEEG records
This study uses separate power and phase feature penalties and the rule that can remove the added block. These rules control the cost of extra features.
- Three forward validation blocks did not show a clear Brier error reduction for separate power and phase penalties in eight fixed SEEG records
[Legacy relation: reproduces] Use the separate penalty engine, candidate grid and one-block rule. All 64 archived one-block metric rows and selected C values match exactly across both tapers. The new test changes the validation schedule and keeps these inputs fixed.
- usesPhase-lag features showed no clear error decrease across sessions for eight-pair SEEG load readers in 15 patients under assumed timing
This earlier test of separate power and phase penalties supplies context for a model limit. Our main test uses one shared penalty. The earlier test used another dataset.