Decoder recalibration

Decoder recalibration is the periodic or continual re-fitting of a decoder to keep performance from decaying as the neural signal changes across hours, days, and months. It is required because the recorded population is nonstationary: the specific units on each channel drift, array micromotion and tissue changes alter waveforms, baseline firing shifts, and the user's own strategy evolves. Without recalibration, a decoder fit on one day typically degrades on the next; with it, chronic systems can be usable for years.

Recalibration methods span a spectrum of supervision. Supervised recalibration collects a short block of known-intent data (e.g. instructed reaches or a cued typing passage) and refits, which is reliable but costs user time. Unsupervised or self-recalibrating decoders instead infer the intended target retrospectively from the user's own corrective behavior during ordinary use — for example, retrospective target inference assumes that once the user selects an item, the movement just before pointed at it — and update parameters without any dedicated calibration block. The practical goal is a system that stays accurate with minimal, ideally invisible, calibration overhead.