Sunday, December 30, 2007

Supervised Motor Learning in the Cerebellum


In the previous article, I made a falsifiable prediction about the cerebellum based on my interpretation of certain metaphorical passages in the book of Revelation. One should note that the current consensus among neurologists regarding cerebellar contribution to speech processing in the brain goes contrary to my prediction. So clearly, nobody can accuse me of using existing scientific literature to make predictions after the fact. Many neurologists have concluded that the cerebellum participates in the production of speech after noticing that patients with cerebellar lesions exhibit speech difficulties. My claim is that they are mistaken and that the cerebellum contributes nothing at all to the generation of speech and language. Its purpose is to attend to routine non-speech-related motor tasks (such as maintaining posture) that would otherwise have to be performed by the motor cortex. I argue that the type of speech impairment observed in patients with cerebellar lesions is due to the motor cortex having to attend to tasks that should normally be handled by the cerebellum. This results in frequent interruptions that manifest themselves as disjointed speech.

One More Cerebellar Prediction From the Bible

The cerebellum is a sensorimotor learning system. Although its learning principles are simple, it can be trained to perform sophisticated sensorimotor tasks such as maintaining posture, walking, running, self-balancing, navigating, etc… Its actions are purely reactive, that is to say, it cannot anticipate the outcomes of sensory or motor patterns. In other words, the cerebellum uses sensory signals to directly control motor effectors in real time.

Two main types of sensory signals are used in the brain. The first is a transient spike or a short spike burst that marks the onset or offset of a sensory phenomenon. The second is a sustained spike train that lasts for the duration of the sensed phenomenon. In the neurobiological literature, these two types of signals arriving at the cortex from the retina are known to go through the magnocellular and the parvocellular pathways. Based on my understanding of the metaphors of Smyrna and Laodicea in the book of Revelation, I can confidently predict that the cerebellum (Laodicea) processes only the second type (sustained spike train) of signals. The Bible uses two metaphors to distinguish between the two: rich (sustained) and poor (transient).

Supervised Learning

In my opinion, the training principle used in the cerebellum is rather simple. It is a trial and error process. Essentially, parallel fibers that receive sensory signals from various places in the body make random synaptic connections with a huge number of Purkinje cells. The output signals generated by a Purkinje cell ultimately activate a muscle. As long as the Purkinje cell is receiving input signals from the parallel fibers, the muscle remains activated. During training, a mature behavior group in the motor cortex (symbolized by the church of Philadelphia) monitors the activation of the muscles under its control and sends a stop signal whenever a muscle is activated longer than it should be. When the stop signal reaches the climbing fiber on the Purkinje cell, a powerful corrective spike is generated. This, in turn, greatly weakens the connections with any parallel fiber that is still firing. Eventually, only the parallel fibers that activate the Purkinje cells at the right time retain their synaptic connections.

Robotic Cerebellum

I think that a simulated software cerebellum can serve as a very effective motor learning system for a humanoid robot. A human trainer using a special sensor-equipped suit that mimics the shape and limbs of the actual robot could teach such a robot to perform various complex tasks such as waking, climbing stairs, etc... Just a thought.

No comments: