[comp.ai.neural-nets] Connectionist simulators for the NeXT?

pbiron@weber.ucsd.edu (Paul Biron) (08/18/90)

[This message cross posted to both comp.sys.next and comp.ai.neural-nets]

Hello all,

Does anyone know of any connectionist simulators for the NeXT?
[I am aware of the Balancer.app demo program and Nexpert Object]

If there aren't any (or even if there are :-), what would others
like to see in a simulator for the NeXT?  I've just bought a cube,
and need a general purpose simulator (akin to SunNet, if anyone
knows that one).  If I can't find one, I'm planning on writting
one of my own (as soon as I get a handle on Objective-C, that is :-),
and would like comments, suggestions, etc from others who would be
interested in such a beast (of course, anything I'd write will
be available to the net).

Here is MY wishlist:

1. Underlying the whole thing would be an Objective-C general
   "network" class, which could be sub-classed by others to add
   in new learning rules, etc. [i.e. "plug-in" learning rules
   and network architectures]

2. This class should have access to the DSP (or maybe Mathematica,
   which has access to the DSP) for its matix operations.

3. This class should be able to execute its operations as multiple
   threads.

4. This class should have access to Mathematica for graphics
   (e.g. 3-D plots of error space).

5. Since my main focus is correlational/Hebbian learning applied
   to Information Retrieval, ability to write the network to disk
   as a Digital Librarian index; and vice-versa. [this one may not
   be very important to anyone else]

6-1000. ?

Anyone who has further ideas, please let me know.

BTW, does anyone know if SunNet has be ported to X yet?  I've only
seen it running in SunView.

Thanx,

Paul Biron      pbiron@ucsd.edu        (619) 534-5758
Central University Library, Mail Code C-075-R
Social Sciences DataBase Project
University of California, San Diego, La Jolla, Ca. 92093