nextuppreviouscontents
Next:Attention and View-Based Object Up:Active Perception Previous:Active Perception

FPGA-Based Spiking Neural Networks

The bulk of digital neural network implementations utilize digital words which encode or represent the average firing rate of the neurons that are being modeled. There is a growing awareness, however, that there is significant information encoded in the relative timing between pulses in a true neural network, and this information is eliminated by the averaging process inherent in computing firing rates. Nerve pulses are perfectly suited to implementation in digital logic, as they are inherently binary. We have been investigating the design of digital circuits that implement various models of spiking neurons, and applying these circuits to the implementation of sensorimotor systems modeled after the human oculomotor system. We have been using the Altera FLEX programmable gate arrays for our implementations. The advantages of using these FPGAs is that it is quick and easy to test and implement in hardware a prototype neural network circuit. Powerful design tools are also readily available for these gate array devices.

J.J. Clark, T.T. Hung


Annual Report

Mon Jun 26 21:22:20 GMT 2000