Amniac--General Purpose Neurocomputing
History:
In the late 1980's, Max Garzon and I, concerned with computabity questions for artificial neural networks, devised a universal neural network analogous to a universal Turing machine. With it we showed that the stability problem was unsolvable for a large class of neural nets (click here and see items 3,12, and 14). Realizing that this universial neural network provided a logical design for a general purpose neurocomputer, we designed one such and called it Amniac. With support from NSF, and with the help of colleagues we implemented Amniac on a massively parallel computer (click here and see item 9). Together with John Caulfield, we also produced a high-level design for an optical implementation of Amniac(click here and see item 15).
Current Status:
With the help of Dave Mundie and others, a hardware implementation of Amniac using field-programmable gate arrays is being prototyped (click here and see item 16).