General Cerebellar Function

The uniformity of the cerebellum's architecture and its well-established role in motor control, including the control of posture and multijoint movements, have motivated theorists to seek a general, unified account of cerebellar function and how it might contribute to motor control and learning. Indeed, given recent anatomical neuroimaging and neuropsychological findings associating the cerebellum and nonmotor functions, these theories have been extended to consider how the cerebellum might provide a computation that is common to a variety of motor and cognitive tasks.

Valentino Braitenberg proposed that the long parallel fibers, through the sequential synapses that they form across a series of Purkinje cells, might constitute a set of delay lines. These delay lines would allow for muscular patterns to be coordinated across synergistic muscle groups recruited in the course of complex actions. Although it appears unlikely that delay lines based on the length and transmission velocity of the parallel fibers can produce a sufficient range of delays, the general idea that the cerebellum is essential for regulating the timing of movements is viable. As noted previously, the cerebellum, especially the cerebellar cortex, is essential for learning when the task requires the precise representation of the temporal relationship between successive events. Indeed, the cerebellum may perform similar timing computations for nonmotor tasks that require this type of representation, providing a general characterization of this system as an internal timing system. Recent computational models indicate that the representation of temporal information may emerge from the complex physiological interactions between the various cells of the cerebellar cortex that shape the output of the Purkinje cells.

Turning to a more mechanistic account of the cerebellum and motor learning, computational models have emphasized that the interactions of the parallel and climbing fibers within the cerebellar cortex can function as a supervised pattern-recognition device. The climbing fibers provide a training signal that modifies the synaptic strength of the connections between the parallel fibers and Purkinje cells. In this manner, the Purkinje cells can shape the topography of movements based on the previous reinforcement history associated with the input patterns provided by the parallel fibers.

Such a mechanism provides an excellent means for learning arbitrary input-output associations but is less obviously useful for computing the temporal relationships necessary for the fine-tuning of motor commands. Researchers, including James Houk and Mitsuo Kawato, have developed this basic model to allow for the dynamic control of the motor system. These models share the properties that critical information about the present and desired state of the effectors, or body parts, is represented along the parallel fibers and that training signals are propagated through the climbing fibers. The initial motor commands originate upstream from the cerebellum. The inhibitory output of the Purkinje cells transforms or sculpts this signal into the appropriate pattern activity to control the actual effector.

Within this framework, it has been proposed that the cerebellum serves as an inverse model of the controlled effector, a forward model, or a combination of the two. A forward model directly simulates the controlled effector, taking as input the motor commands issued to the effector and producing as output the predicted sensory feedback. One might question the use of a forward model in learning since it merely approximates the actual behavior of the controlled effector. However, sensory feedback is quite slow in relation to the speed at which movements are performed and thus, at least during ballistic movements, may be of little help during execution. A model of the effector that can anticipate the behavior of the body in response to motor commands provides one way in which feedback can be rapidly used to contribute to on-line control. The forward model can be used to anticipate the sensory consequences of motor commands, and the difference between the expected and actual sensory feedback can be used as a sort of error signal.

An inverse model, as its name suggests, performs the inverse of the computation performed by the forward model. It takes as input the present and the desired state of the controlled effector and produces as output the motor command required to achieve the desired state. Such a computational device is of obvious use: With an accurate inverse model, a system need only compute the desired trajectory for the effector and allow the model to determine the necessary motor command.

Learning to encode a forward model is straightforward: The system takes as input the present state of the effector and the motor command and learns to produce as output the sensory feedback. Learning to encode an inverse model is much more difficult because the appropriate motor command is not available to the system (if it were, then there would be little need to learn an inverse model). For an inverse model to be correctly trained, observable sensory errors must be converted into motor errors. Therefore, inverse models are comparatively computationally expensive.

Mitsuo Kawato and colleagues proposed that the cerebellum houses the neural circuitry supporting both forward and inverse models. Evidence for the presence of each type of model has come from analyses of the patterns of neural activity within the cerebellum. Such evidence is indirect, however, and researchers continue to strive to characterize the computation performed by the structure.

Understanding And Treating Autism

Understanding And Treating Autism

Whenever a doctor informs the parents that their child is suffering with Autism, the first & foremost question that is thrown over him is - How did it happen? How did my child get this disease? Well, there is no definite answer to what are the exact causes of Autism.

Get My Free Ebook


Post a comment