BMI skill learning and cortical plasticity

BMI skill learning refers to the user's side of the loop: with a fixed decoder held constant, the brain can learn to control the interface as a genuine motor skill, improving over sessions and forming a stable neural representation for the task. Studies holding the decoder constant for days have shown that neurons develop consistent, reproducible tuning to the BCI task, that performance consolidates and is retained across days like a learned skill, and that this map can coexist with the native motor repertoire. This is direct evidence that the cortex treats the artificial effector as a controllable extension of the body.

Skill learning is the complement to decoder adaptation. When the decoder keeps changing (CLDA), the user faces a moving target and cannot fully consolidate a strategy; when the decoder is fixed, the user can learn but is stuck with whatever quality that decoder allows. Understanding which changes should be assigned to the machine and which to the brain — and over what timescales the two learners should each move — is the core scientific question behind co-adaptive BCI design. Constraints matter too: users learn readily within the existing neural manifold but struggle to produce activity patterns off that manifold, bounding what a fixed decoder can teach.