pography ( Clark, McCormick, Lavond, & Thompson, 1984). Although we know of no direct evidence for the proposed cerebellar implementation of the VET model, particularly regarding the associative depression of PF/Golgi cell synapses, we believe that it is at least physiologically plausible.
Despite these successes, it is possible that the output of PCs that are the model's V-units would be insufficient to generate CRs.11 Were this the case, the cerebellar VET network may only be required to construct a template of the temporally adaptive CR. This template could be fed back to cerebellar cortex through collaterals from nuclear cells ( Haines, 1988). This feedback would provide input to a neural network capable of generating CRs. Such a network might resemble the one proposed by Moore and Blazis ( 1989) to implement the SBD model.12
Levey and Martin ( 1968) suggested that the complexities of learning a temporally adaptive CR lend credence to multistage approaches such as the three- state Markov model proposed by Theios and Brelsford ( 1966). Prokasy ( 1987) makes a similar point in support of two-stage models of CR acquisition. A two- component cerebellar network might resolve the dilemmas posed by Berthier and Moore ( 1986) study of CR-related activity by HVI PCs during NMR conditioning. The firing pattern of the majority of CR-related PCs in the Berthier and Moore ( 1986) study can be accounted for by the Moore and Blazis' ( 1989) implementation of the SBD model and not by the VET implementation. Berthier and Moore ( 1986) observed only one PC in HVI with a firing pattern in complete accord with this scheme. It could well be the case that only a few PCs are required to generate a CR template, but many more acting in concert, as portrayed in the SBD network, are required for long-term storage of learning and the generation of robust CRs. Moore and Blazis ( 1989) discuss ways in which other brain systems (e.g., those involving the hippocampus) might be used to generate temporally adaptive templates for the SBD processing unit. The output of the VET network has this capability.
The authors thank Diana Blazis for helpful comments on the manuscript and Neil Berthier, Diana Blazis, James Houk, and William Richards for their expertise on various aspects of the proposed cerebellar implementation of the VET model. Preparation of this chapter was supported by grants from the AFOSR and NSF.____________________