Abstract
The hardware, which can be associatively recall information as does a brain, is one of the attractive applications of neural networks. Here a new type of optical neural network system, which can simulate the brain’s intelligent associative ability, is proposed. The combination of both the associative ability of neural networks and the logical decision-making of electrical computer is achieved through a modification of the system cost function. In our experiment, a work-reader system is built by introducing human knowledge—a dictionary—which makes the system recall a meaningful combination of patterns. A fully interconnected optical neural network with 7×7 input neurons and 7×7 output neurons is implemented by LCTV to memorize the alphabetical patterns. Dictionary information is introduced into the network by changing the threshold levels of the output neurons dynamically. The experiment shows that the system possesses almost the same recognition ability as a brain even when a low dynamic range of LCTV is used as the key hardware for interconnection. Most of the local minima caused by the hardware inhomogeneity, inaccuracy or maybe the imperfectness of the memory model, can be effectively avoided. The whole system is now being integrated on a VLSI chip in Mitsubishi Electric Corporation for practical uses.
© 1991 Optical Society of America
PDF ArticleMore Like This
K.J. Weible, N. Collings, W. Xue, G. Pedrini, and R. Dändliker
ME9 Optical Computing (IP) 1991
Mark A. Neifeld and Demetri Psaltis
WA4 Optical Computing (IP) 1991
Lynda M. Ralston, William R. Beaudet, and Reeder N. Ward
MW.4 OSA Annual Meeting (FIO) 1993