home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!spool.mu.edu!yale.edu!ira.uka.de!Germany.EU.net!infko!inews
- From: evol@infko.uni-koblenz.de
- Subject: Re: Integer neural networks?
- Message-ID: <1992Dec14.125135.24524@infko.uucp>
- Sender: inews@infko.uucp (inews)
- Organization: University of Koblenz, Germany
- References: <R5BFVB4w165w@west.darkside.com>
- Date: Mon, 14 Dec 1992 12:51:35 GMT
- Lines: 24
-
- In article <R5BFVB4w165w@west.darkside.com> max@west.darkside.com (Erik Max Francis) writes:
- >Anyone know of any good schemes (or books which describe schemes) of
- >neural networks that deal with only integer math?
- >
- >I'm looking more to use the neural network as a "black box" that receives
- >certain inputs, does manipulations, and then spits out certain outputs.
- >Finite state machines would work fine, too, but there are too many
- >(potential) inputs to implement reasonably.
- >
-
- Hallo !
-
- Some weeks ago Donald Tveter posted a backprop implementation to
- comp.sources.misc, which can use both integer and float arithmetic.
- If have compiled it for NeXTstation and SPARC2. There is no need to
- use integer arithmetic on these machines. Floating point arithmetic
- is as fast as integer arithmetic or even faster if you are using a
- fast Coprozessor. Nevertheless integer arithmetic is fine for machines
- without an arithmetic Coprozessor. If you want a hardware implementation
- of your ANN you have to use integer arithmetic in most cases.
-
-
- Randolf Werner
-
-