This chapter consists of a paper presented at the fifth Neural Information Processing Systems conference in December 1992. Based on the monkey VOR models by Miles et al. (miles80) and Lisberger (Lisberger88a), the goal was to study the VOR adaptation hypothesis advanced by Miles & Lisberger (mileslis81). Their hypothesis had two parts: 1) that long-term adaptation occurs in the brainstem vestibular pathways and not in the cerebellum; 2) that the cerebellum has an inductive role in the adaptation, providing at least part of the error signal guiding the long-term adjustments in the brainstem via the Purkinje cells of the cerebellum.
Their hypothesis was verified under certain constraints on the adaptation parameters. That is, in order to abolish an overshoot in the VOR gain which is not observed during learning in monkeys, the following was necessary: 1) adaptation in the brainstem must be driven mainly by visual error reported by climbing fiber inputs from the inferior olive, and to a much lesser extent by the Purkinje cell signal; and 2) the learning rate of adaptation in the cerebellum must be much greater than the learning rate in the brainstem. The results were obtained analytically and verified with computer simulations.
The model suggests how the dynamics of the VOR system can lead to long-term adaptation which differs from what may be expected from local learning rules at the synapses because of differences in time scales and shifts of activity in the system during learning. This may reconcile apparently contradictory evidence between local learning rules observed in vitro in the flocculus ([Ito1970]) and the long-term adaptation seen in vivo in animals (Miles and Lisberger, 1981).