Using these tasks we demonstrate that cerebellar feedback predictions conveyed to the cerebral cortex facilitate learning. We test our model on a range of sensorimotor, pattern recognition and visual-language tasks. ![]() generates cerebral feedback predictions) given current cerebral activity. This view of cerebro-cerebellar interactions is in line with the classical forward models of cerebellar function 6, 7, in that in our model the cerebellum makes forward predictions (i.e. Specifically, we model a given cerebral area as a recurrent neural network 27, 28, 29, 30 which receives feedback predictions from a feedforward, cerebellar, network 6, 7. This feedback predicted by the cerebellum is then sent back to the cerebral network to drive learning. Building on recent deep learning developments we theorise that the cerebellum predicts future cerebral feedback signals given current cerebral activity. Taken together, these studies suggest that the cerebellum learns internal models for both motor and non-motor functions in line with the proposed universal functional role of the cerebellum across the brain, including the cerebral cortex 9, 24, 25, 26.ĭespite growing experimental evidence there are no specific computational models aiming to capture the functional roles of cerebro-cerebellar interactions during learning of motor and non-motor tasks. An increasing body of behavioural 12, 14, 16, 17, 18, 19, 20, anatomical 21, 22 and imaging 23 studies allude to a role of the cerebellum in cognition in animals and humans. However, more recently, cerebellar dysfunction has also been associated with impaired language processing, cognitive associative learning and working memory 11, 12, 13, 14, 15. Consistent with this view are a large body of experimental observations for which cerebellar dysfunction causes motor learning deficits. In the classical view, the cerebellum learns predictive internal models on the motor domain 5, 6, 7, 8, 9, 10. The cerebellum is a region of the brain specialised in building predictive models 4, 5. These observations suggest that the brain may employ a general mechanism to facilitate learning when external feedback is not readily available. However, external sensory feedback is inherently delayed and incomplete, thereby reducing the rate and extent of learning in neuronal circuits 3. To learn efficiently animals and humans must make good use of this feedback to update their internal models of the world 3, 4. Learning ultimately depends on environmental feedback 1, 2. Overall, our work offers a theoretical framework of cerebro-cerebellar networks as feedback decoupling machines. ![]() Finally, the model makes several experimentally testable predictions regarding cerebro-cerebellar task-specific representations over learning, task-specific benefits of cerebellar predictions and the differential impact of cerebellar and inferior olive lesions. Next, we demonstrate that these results generalise to more complex motor and cognitive tasks. When trained in a simple sensorimotor task the model shows faster learning and reduced dysmetria-like behaviours, in line with the widely observed functional impact of the cerebellum. In this model a cerebral recurrent network receives feedback predictions from a cerebellar network, thereby decoupling learning in cerebral networks from future feedback. Inspired by recent deep learning algorithms, we introduce a systems-level computational model of cerebro-cerebellar interactions. How the cerebral cortex learns efficiently despite the sparse nature of feedback remains unclear. However, such feedback is often not readily available. Behavioural feedback is critical for learning in the cerebral cortex.
0 Comments
Leave a Reply. |