Neutral Network Models of Primate Motor System.
Final technical rept.,
WASHINGTON UNIV SEATTLE DEPT OF PHYSIOLOGY AND BIOPHYSICS
Pagination or Media Count:
The primary goal of this project was to develop neural network models of the primate sensorimotor system, and we succeeded beyond our original expectations. Over the two-year span of this project we acquired the requisite hardware a SUN 3-260 and two NeXT computers, developed a unique dynamic recurrent neural network program called NeXTNet, and derived neural network simulations for several primate sensorimotor tasks. NeXTNet is a dynamic recurrent computer network simulator based on the algorithms of Watrous 1986 and Williams and Zipser 1989 it was designed to facilitate easy set-up of networks and gradient descent backprop training of a wide range of spatiotemporal transforms. Being dynamic, the model can incorporate as target patterns the firing patterns of neurons and motor units previously record in behaving monkeys. The network is also fully recurrent, allowing unrestricted connectivity, so the circuitry can be made to resemble anatomical pathways with feedback and collateral connections. The inputs to the network were step changes in position of a visual target, and the outputs were the eight discharge patterns of flexor and extensor motor units.
- Computer Hardware