cs178aas@sdcc8.ucsd.EDU (Grobbins) (02/23/88)
A few weeks ago I posted a request for information about any neural net implementations using HyperCard. Several people asked for a summary of what turned up; no one, with one exception, knew of existing, appropriate code. Chris Borton did have a copy of Larry Halff's Neural Net Construction Stack. NNCS appears to be a rather crude attempt at neural net simulation, in which each processing element is a separate card, with fields holding link information, weights, and the PE's transfer function. NNCS includes a ten-cell retina simulation, a primitive on-center/off-surround net. I didn't play around with NNCS much, since it is limited by the power of interpreted HyperTalk. I have a backprop net written in LSP which, if time makes itself available in April, I may try to turn into a Hyper- Card XFCN. This would put a straightforward front end on an otherwise not especially inviting network. If you have knowledge of any existing, useful integrations of neural nets and HyperCard, please tell the net about them. Grobbins grobbins%sdemlab@ucsd.edu ..ucsd!sdemlab!grobbins