ErrP-based correction and adaptation

ErrP-based adaptation uses decoded error-related potentials as an internally generated supervisory signal, so a BCI can improve without external labels. In its simplest correction form, a detected ErrP after an action causes the system to reject or reverse that action; in its learning form, the ErrP relabels the just-made decision as wrong and drives a parameter update, effectively giving the decoder access to the user's own error monitoring. Because the teaching signal is the user's brain response rather than an experimenter's key, this enables self-calibration during ordinary use and connects naturally to reinforcement-style formulations in which the ErrP acts as a negative reward.

The method inherits the frailties of ErrP decoding. Detection errors propagate: a missed ErrP leaves a real mistake uncorrected, and a false ErrP corrupts a correct action or injects a bad label into the update. Because these labels then feed adaptation, the scheme can enter a self-reinforcing failure mode if ErrP reliability degrades. Practical systems therefore weight ErrP-derived labels by detection confidence, combine them with other evidence, and limit how much any single inferred error can move the decoder.