rao@cadillac.CAD.MCC.COM (Arun Rao) (07/07/88)
This is the first time I'm posting anything on this net, so I was not able to retrieve the address of the person who wanted to know about existing text recognition software using neural nets. I have a program which implements Grossberg's ART1 (with some modifications) for text recognition. It is still rudimentary, but is capable of adaptively learning to recognize a bit-mapped set of characters at its input. Access after learning is virtually instantaneous, and the number of characters it can store depends on how many connections you can simulate in your machine. I have tried the program with a large number of input nodes and there are no problems. The time taken by the network to learn (I refer to real time, as in the time taken if the network were to be built in hardware) does not scale up to any great degree with the number of nodes (CPU time obviously does). Arun Rao Arizona State University (Currently a summer intern at MCC) -- Arun Rao @ MCC VLSI CAD Program [512] 339-3793 P.O. Box 200195, Austin, TX 78720 ARPA: rao@mcc.com UUCP: {ihnp4,seismo,harvard,gatech,pyramid}!ut-sally!im4u!milano!bell!rao