Trends in Cognitive Sciences
Internal models in the cerebellum
Section snippets
The cerebellum as an inverse model
Fast and coordinated arm movements cannot be executed under pure feedback control because biological feedback loops are both too slow and have small gains. Two major feedforward control schemes have been proposed: the equilibrium-point control hypothesis6, 7, 8, 9and the inverse dynamics model hypothesis[4]. Some versions of the former scheme advocate that the central nervous system (CNS) can avoid complicated computations by relying on the spring-like properties of muscles and reflex loops.
The cerebellum as a forward model
An alternative hypothesis for the cerebellum, proposed under a variety of forms32, 33, 34, 35, suggests that the cerebellum generates a forward, causal representation of the motor apparatus, often known as a forward model5, 36, 37. A forward model represents the normal behavior of the motor system in response to outgoing motor commands. Hence, a forward model of the arm's dynamics has, as inputs, the current state of the arm and an efferent copy of motor commands being issued by a controller,
The cerebellum as multiple paired forward and inverse models
The previous sections have focused on the utility and evidence for either inverse or forward models within the cerebellum for generation of motor commands and control of movements. In this section, we speculate on the benefits of multiple internal models and, in particular, the advantages of pairing inverse and forward models for motor learning and control. Although the cerebellum has often been viewed as a modular system3, 55, we present a new cohesive computational framework for motor
Conclusion
Internal models provide a firm computational foundation from which theories of the cerebellum can be considered. We have reviewed the evidence that the cerebellum contains inverse or forward models of the motor system. By considering the possibility that the cerebellum contains multiple pairs of forward and inverse models, we believe that the benefits of both views can be retained and integrated. Such a paired system would results in computational advantages in both motor learning and control.
Outstanding questions
- •
There is evidence that the cerebellum contributes to an inverse dynamics model of the eye. But similar evidence for the hand and arm is less clear. Does this imply different function for the cerebellum in controlling different effectors or does it simply reflect the different levels of complexity, making identification of internal models of the hand less certain?
- •
Are the different types of model (forward and inverse) found in different areas of the cerebellar cortex? Are the paired models
Acknowledgements
This work was supported by grants from the Wellcome Trust, the Medical Research Council, the Royal Society, the BBSRC and the Human Frontier Science Project. R.C.M. is supported by a Wellcome Senior Research Fellowship.
References (75)
A theory of cerebellar function
Math. Biosci.
(1971)- et al.
Forward models: supervised learning with a distal teacher
Cognit. Sci.
(1992) - et al.
The cerebellum and VOR/OKR learning models
Trends Neurosci.
(1992) Properties of cerebello–precerebellar reverberating circuits
Brain Res.
(1983)Cerebellar learning in vestibulo–ocular reflex
Trends Cognit. Sci.
(1998)Expression of a protein kinase-C inhibitor in Purkinje cells block cerebellar LTD and adaptation of the vestibulo–ocular reflex
Neuron
(1998)Computational approaches to motor control
Trends Cognit. Sci.
(1997)- et al.
Forward models for physiological motor control
Neural Netw.
(1996) The relevance of sensory input for the cerebellar control of movements
NeuroImage
(1997)- et al.
Motion perception deficits from midline cerebellar lesions in human
Vision Res.
(1995)