A half-weight update raised mean relative forecast error by 1.0% in selected SEEG readouts from eight sessions after contact order reversal

We tested whether one new stimulation response could improve a stored SEEG response forecast. We used public STReEF data from eight people with epilepsy. Five responses at the original contact order built the forecast. The first response at the reversed order supplied a fixed half-weight update. Four later responses were held out. The test covered 668 instances of recording pairs at selected conditions from 20 to 500 ms after stimulation. Their original responses passed a stability check. Error fell in seven people. A large increase in one person gave a mean relative gain of −1.0%. The 95% person bootstrap interval was −37.6% to 23.8%. The result does not show a mean benefit. Across all 6,803 instances, error rose in every person. A two-pulse update gave a 17.5% mean gain in an exploratory test. Contact order and time were linked. This study cannot separate their effects. The package has three figures, source checks, exact response products, an offline forecast tool, and reproduction code.

stereoelectroencephalography · electrical stimulation · brain-computer interfaces · forecasting · signal processing · epilepsy