User learning and neural plasticity in closed loop

User learning is the biological half of co-adaptation: with practice under feedback, cortical activity reorganizes so that the user drives a given decoder better over time. This is fundamentally operant — the user learns, largely without explicit instruction, to produce whatever neural patterns the decoder rewards with good control — and it can proceed even when the decoder is held completely fixed, ruling out decoder adaptation as the sole source of improvement. Timescales range from within-session gains over minutes to consolidation and refinement across days and weeks.

A key structural finding is that this learning is not unconstrained. Users acquire new neural activity patterns much more readily when those patterns lie within the pre-existing low-dimensional neural manifold of correlated population activity than when the required patterns fall outside it, so the intrinsic geometry of cortical activity shapes what is easy to learn to control. For closed-loop design this is decisive: the user is the biological learner the two-learner problem must accommodate, and giving that learner a stable decoder to practice against is often what lets a durable, consolidated neural map — and expert-level control — emerge.