Adaptive decoding

Adaptive decoding is the broad category in which decoder parameters change over time rather than being fixed after calibration. The motivation is that neural signals are non-stationary on many timescales — electrode micromotion and impedance change over days, arousal and fatigue over minutes, and the user's own learning throughout — so a decoder frozen at calibration will drift out of alignment with the signals it must interpret. Adaptation methods are usually classified by their supervision: supervised (true labels available from instructed tasks), semi-supervised (labels estimated from context or decoder confidence), and unsupervised (tracking distribution shift without labels at all).

The unavoidable danger is that adaptation forms a second feedback loop on top of the control loop. If the decoder adapts on labels that are themselves derived from its own outputs, errors can reinforce themselves; if it adapts too aggressively, it couples with the user and undermines the stable mapping the user is trying to exploit. Sound adaptive decoding therefore pairs an adaptation rule with safeguards — rate limits, confidence gating, drift detection, or periodic reversion to a trusted reference — rather than adapting unconditionally.