Feedback-based motor control can guide plasticity and drive rapid learning

Abstract

Animals use afferent feedback to rapidly correct ongoing movements in the presence of a perturbation. Repeated exposure to a predictable perturbation leads to behavioural adaptation that counteracts its effects. Primary motor cortex (M1) is intimately involved in both processes, integrating inputs from various sensorimotor brain regions to update the motor output. Here, we investigate whether feedback-based motor control and motor adaptation may share a common implementation in M1 circuits. We trained a recurrent neural network to control its own output through an error feedback signal, which allowed it to recover rapidly from external perturbations. A biologically plausible plasticity rule based on this same feedback signal allowed the network to learn to counteract persistent perturbations through a trial-by-trial process, in a manner that reproduced several key aspects of human adaptation. Moreover, the resultant network activity changes were also present in neural population recordings from monkey primary motor cortex. Online movement correction and longer-term motor adaptation may thus share a common implementation in neural circuits.

Feedback control of recurrent dynamics constrains learning timescales during motor adaptation

“…In fact, our results using weight perturbation for retraining inputs were inspired by previous work on slow, noisy cerebellar-dependent adaptation [89, 90]. Alternative computational theories also suggest that cerebellar feedback may guide subsequent local plasticity within M1, either by acting as a teaching signal [91, 92] or by estimating surrogate gradients for weight updates [93]. Nonetheless, better characterisation of cortico-cerebellar interactions during BCI control as well as learning is a promising next step.…”

“…While we exclusively modelled upstream plasticity as the basis of adaptation, it is likely there are concurrent distributed changes, particularly for longer- timescale learning. Several computational studies have suggested error- or reward-based modulation of local connectivity in M1 during motor adaptation [92, 100, 101]. Modelling these parallel learning processes and dissecting the multi-region circuits underlying learning is an exciting avenue for future work.…”