home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!ulowell!m2c!jjmhome!schunix!sonix
- From: sonix@schunix.uucp (Duane Morin)
- Newsgroups: comp.ai
- Subject: "Mini" Neural Net Application: Ref's?
- Message-ID: <1992Aug28.202942.27516@schunix.uucp>
- Date: 28 Aug 92 20:29:42 GMT
- Organization: SCHUNIX Public Access Unix for Worcester County, Massachusetts, USA
- Lines: 34
- X-Newsreader: Tin 1.1 PL4
-
- Ok, so we have an application at work that I think would benefit from the
- application of a simplified neural net approach (I get this train of thought
- from the C Users Journal, Sept 1992, article of the same theme). Here's
- what I figure:
-
- We have an application that calculates a "baseline" of 12 points over
- a series of increasing frequencies. This data is then graphed, and
- it is up to the user to decide whether or not the trace is acceptable,
- according to the general shape of the trace drawn. The pictures that
- the user sees will be fairly clear: OK trace, Input Cable Not
- Working (flat line), Object in Path (Crooked line), etc... *BUT*
- each trace will be different, i.e. two OK traces will not be IDENTICAL
- but will look basically the same.
-
- Now, the question. Does anyone more familiar with the topic than I think
- that we can implement something to analyze these traces and predict what
- class they fall in, without having to go to the user? Of course, the user
- could be consulted as a last ditch effort, i.e. if the program could not
- recognize the trace as belonging to a particular class.
-
- I was thinking that maybe we could take several examples of each kind of
- trace and "train" the computer to recognize them. That way, when the user
- creates a trace of a particular class, the computer would be able to
- distinguish between the various (say, 10 ) classes of traces.
-
- Is this possible? Feasible? Thanks for any responses. Email preferred,
- summary to the net if there's interest.
-
- Thank you,
- Duane Morin
- Software Development Manager
- Walker Sonix
- Worcester, MA 01604
- sonix@schunix.uucp
-