A NEURAL-NET MODEL OF THE ADAPTATION OF BINOCULAR VERTICAL EYE ALIGNMENT

Binocular eye alignment is continuously recalibrated and readjusted to maintain a single view of the world. Once this process is complete, v isual feedback is no longer required to maintain alignment. Rather, al ignment is maintained through non-visual or extra-retinal information. The calibration process can be demonstrated by producing a cross-coup ling or association between vertical vergence and another type of eye movement. This paper presents a neural net model of a plausible biolog ical mechanism that could be involved with maintaining alignment in th e context of vertical vergence. The model couples conjugate eye-positi on-sensitive neurons with a vertical vergence response. Weight trainin g of the input neurons is accomplished with a modified Hebbian rule th at minimizes the vertical eye alignment error during adaptation to ver tical disparities. The experimental results are simulated with a class of input neurons that has randomly distributed sensitivities and thre sholds similar to those found in premotor sites in the brainstem. For simultaneous adaptation to three vertical disparities, the weighted in puts of the input class are reshaped such that the inputs qualitativel y obtain a sensitivity-threshold relationship similar to that of moton eurons in the brainstem.