The Back Propagation Technique for Modeling Cortical Computation
Abstract
Over the past several years powerful learning procedures have been developed that can program simulated neural networks to compute a wide variety of functions. This has made it possible to use learning procedures to train model networks to do computations that occur in the brain. While there was so a priori reason to suppose that the individual neuro-like units in these model networks would resemble the brain in any way, the empirical observations is that they do. Good results have been achieved applying this paradigm to modeling monkey parietal area 7a. Various aspects of the primary visual area have also been successfully modeled using this approach. The results of this work raise the interesting possibility that learning procedures and particularly the back propagation algorithm used in these studies, can serve as a general technique to account for how the brain implements computations. While these observations do not imply that back propagation is actually used in the brain, they do raise the possibility that some analogous learning procedure is used there.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 31, 1989
- Accession Number
- ADA209855
Entities
People
- David Zipser
Organizations
- University of California, San Diego