Probabilistic models of state estimation predict visuomotor transformations during prism adaptation

Bayesian State Estimation in Sensorimotor Systems With Particle Filtering

In sensorimotor control, sensory feedback integrates with forward models to alleviate the impacts of sensory noise and delay on state estimation. The sensorimotor integration is subject to Bayesian inference and has been formulated by the Kalman filter in computational neuroscience. However, the Kalman filter, as an artificial optimal estimator to address the abstract characteristics of spatial perception, is inadequate to present the neural computation in the cerebellum. Besides, the nonlinear neuromuscular dynamics with tightly coupled state variables also substantially impedes the implementation of Kalman filter in realistic sensorimotor systems. Here we address the sensorimotor state estimate by using the particle filter, a nonlinear Bayesian estimator that can be implemented in arbitrary dynamic systems with the neurocomputational compatibility. Particle filtering is explicitly implemented in a biophysically realistic sensorimotor model of an upper limb integrating Hill-type muscles, tendons, skeleton, and primary afferents. By involving the command noises, the constructed neuromusculoskeletal model qualitatively represents the experimental variability in center-out reaching movements. Despite the initial estimation uncertainty and sensorimotor noises, the particle filter is able to approximate the actual states in forward-reaching movements. Furthermore, the simulated hand-position estimate is consistent with the experimental results, in the presence of forward model errors, neural noises, and sensory delays. The particle filter is demonstrated to effectively implement the Bayesian state estimation in biophysically realistic sensorimotor systems and provide better compatibility with neuronal computation than the Kalman filter.

Tandem internal models execute motor learning in the cerebellum

In performing skillful movement, humans use predictions from internal models formed by repetition learning. However, the computational organization of internal models in the brain remains unknown. Here, we demonstrate that a computational architecture employing a tandem configuration of forward and inverse internal models enables efficient motor learning in the cerebellum. The model predicted learning adaptations observed in hand-reaching experiments in humans wearing a prism lens and explained the kinetic components of these behavioral adaptations. The tandem system also predicted a form of subliminal motor learning that was experimentally validated after training intentional misses of hand targets. Patients with cerebellar degeneration disease showed behavioral impairments consistent with tandemly arranged internal models. These findings validate computational tandemization of internal models in motor control and its potential uses in more complex forms of learning and cognition.

Towards a neuro-computational account of prism adaptation

Prism adaptation has a long history as an experimental paradigm used to investigate the functional and neural processes that underlie sensorimotor control. In the neuropsychology literature, prism adaptation behaviour is typically explained by reference to a traditional cognitive psychology framework that distinguishes putative functions, such as 'strategic control' versus 'spatial realignment'. This theoretical framework lacks conceptual clarity, quantitative precision and explanatory power. Here, we advocate for an alternative computational framework that offers several advantages: 1) an algorithmic explanatory account of the computations and operations that drive behaviour; 2) expressed in quantitative mathematical terms; 3) embedded within a principled theoretical framework (Bayesian decision theory, state-space modelling); 4) that offers a means to generate and test quantitative behavioural predictions. This computational framework offers a route towards mechanistic neurocognitive explanations of prism adaptation behaviour. Thus it constitutes a conceptual advance compared to the traditional theoretical framework. In this paper, we illustrate how Bayesian decision theory and state-space models offer principled explanations for a range of behavioural phenomena in the field of prism adaptation (e.g. visual capture, magnitude of visual versus proprioceptive realignment, spontaneous recovery and dynamics of adaptation memory). We argue that this explanatory framework can advance understanding of the functional and neural mechanisms that implement prism adaptation behaviour, by enabling quantitative tests of hypotheses that go beyond merely descriptive mapping claims that ‘brain area X is (somehow) involved in psychological process Y’.

•

Traditional neuropsychological models of prism adaptation lack precision.

•

Computational models improve explanatory and predictive power.

•

A range of adaptation phenomena can be explained quantitatively.

•

Mathematics offers a bridge between neural mechanisms and behaviour.

•

A neuro-computational approach will advance neuropsychology.

Dynamic neural correlates of motor error monitoring and adaptation during trial-to-trial learning

A basic EEG feature upon voluntary movements in healthy human subjects is a β (13-30 Hz) band desynchronization followed by a postmovement event-related synchronization (ERS) over contralateral sensorimotor cortex. The functional implications of these changes remain unclear. We hypothesized that, because β ERS follows movement, it may reflect the degree of error in that movement, and the salience of that error to the task at hand. As such, the signal might underpin trial-to-trial modifications of the internal model that informs future movements. To test this hypothesis, EEG was recorded in healthy subjects while they moved a joystick-controlled cursor to visual targets on a computer screen, with different rotational perturbations applied between the joystick and cursor. We observed consistently lower β ERS in trials with large error, even when other possible motor confounds, such as reaction time, movement duration, and path length, were controlled, regardless of whether the perturbation was random or constant. There was a negative trial-to-trial correlation between the size of the absolute initial angular error and the amplitude of the β ERS, and this negative correlation was enhanced when other contextual information about the behavioral salience of the angular error, namely, the bias and variance of errors in previous trials, was additionally considered. These same features also had an impact on the behavioral performance. The findings suggest that the β ERS reflects neural processes that evaluate motor error and do so in the context of the prior history of errors.