home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!gatech!hubcap!luna!bdbryan
- From: bdbryan@eng.clemson.edu (Ben Bryant)
- Subject: Re: Neural Net for Matlab
- Message-ID: <1992Jul23.014930.19522@hubcap.clemson.edu>
- Sender: news@hubcap.clemson.edu (news)
- Reply-To: bdbryan@eng.clemson.edu
- Organization: College of Engineering, Clemson Univ.
- References: <35984@sdcc12.ucsd.edu>
- Date: Thu, 23 Jul 1992 01:49:30 GMT
- Lines: 35
-
- In article 35984@sdcc12.ucsd.edu, demers@cs.ucsd.edu (David DeMers) writes:
- >In article <1992Jul22.134759.25636@uceng.UC.EDU> chan@uceng.UC.EDU (Alistair Keating Chan) writes:
- >
- >>Has anyone seen or written a standard backpropgation algorithm
- >>for Matlab that they would like to and be able to share?
- >
- >No, but MathWorks sells a Neural Networks Toolbox, just out. I don't
- >know the price. I saw it demonstrated a couple of months
- >ago and it seems pretty spiffy. Sorry, I don't have a
- >better or more detailed review than that :-)
- >
- >However, standard bp is pretty easy to code up, in my opinion.
- >Forward activation is just matrix multiply, squash, matrix multiply,
- >squash...
- >and standard delta rule learning is
- >compute delta, update weights, back prop (matrix mult),
- >compute delta, update weights, back prop...
- >
- >Dave
- >
- >--
- >Dave DeMers ddemers@UCSD demers@cs.ucsd.edu
- >Computer Science & Engineering C-014 demers%cs@ucsd.bitnet
- >UC San Diego ...!ucsd!cs!demers
- >La Jolla, CA 92093-0114 (619) 534-0688, or -8187, FAX: (619) 534-7029
-
- Close.
-
- compute delta, back prop (if more than one hidden layer),
- compute delta, back prop (through all layers)...
- then update weights
-
- -B.
-
-
-