Benefit of additional calibration data under matched optimization updates

Determine whether calibration data beyond three minutes improves sEMG speech-decoding performance when the compared calibration budgets receive matched numbers of optimizer updates.

Background

The calibration experiment compared one-, three-, five-, ten-minute, and approximately 13-minute budgets using equal epoch counts rather than equal numbers of optimizer updates. Larger calibration pools therefore produced more updates per epoch, confounding calibration duration with optimization exposure. The authors explicitly leave unresolved whether additional labeled calibration data would improve performance under a schedule that controls for the number of updates.

References

The experiment does not establish equivalence among the 3-, 5-, 10-min, and full budgets, which were not compared with one another, nor does it show whether additional data would be beneficial under a matched-update schedule.

— Multi-Subject Pretraining Enables Short-Calibration Personalization for Closed-Corpus Surface EMG Speech Decoding  (2609.21288 - Le et al., 18 Sep 2026) in Section 4.3, “Three Minutes of Calibration Yielded Similar Error Rates to Full Calibration”