THE OCULOMOTOR INTEGRATOR - TESTING OF A NEURAL-NETWORK MODEL

An important part of the vestibule-ocular reflex is a group of cells i n the caudal pens, known as the neural integrator, that converts eye-v elocity commands, from the semicircular canals for example, to eye-pos ition commands for the motoneurons of the extraocular muscles. Previou sly, a recurrently connected neural network model was developed by us that learns to simulate the signal processing done by the neural integ rator, but it uses an unphysiological learning algorithm. We describe here a new network model that can learn the same task by using a local , Hebbian-like learning algorithm that is physiologically plausible. T hrough the minimization of a retinal slip error signal the model learn s, given randomly selected initial synaptic weights, to both integrate simulated push-pull semicircular canal afferent signals and compensat e for orbital mechanics as well. Approximately half of the model's 14 neurons are inhibitory, half excitatory. After learning, inhibitory ce lls tend to project contralaterally, thus forming an inhibitory commis sure. The network can, of course, recover from lesions. The mature net work is also able to change its gain by simulating abnormal visual-ves tibular interactions. When trained with a sine wave at a single freque ncy, the network changed its gain at and near the training frequency b ut not at significantly higher or lower frequencies, in agreement with previous experimental observations. Commissural connections are essen tial to the functioning of this model, as was the case with our previo us model. In order to determine whether a commissure plays a similar r ole in the real neural integrator, a series of electrical perturbation s were performed on the midlines of awake, behaving juvenile rhesus mo nkeys and the effects on the monkeys' eye movements were examined. Eye movements were recorded using the coil system before, during, and aft er electrical stimulation in the midline of the pens just caudal to th e abducens nuclei, which reversibly made the integrator leaky. Eye mov ements were also recorded from two of the monkeys before and after a m idline electrolytic lesion was made at the location where stimulation produced a leaky integrator. This lesion disabled the integrator irrev ersibly. The eye movements that were produced by the monkeys as a resu lt of these perturbations were then compared with eye movements produc ed by the model after analogous perturbations. The results are compati ble with the hypothesis that integration comes about by positive feedb ack through lateral inhibition effected by an inhibitory commissure.