home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!zaphod.mps.ohio-state.edu!cs.utexas.edu!asuvax!ncar!noao!amethyst!organpipe.uug.arizona.edu!hindmost!jnelson
- From: jnelson@lpl.arizona.edu (Jeff Nelson)
- Newsgroups: rec.games.netrek
- Subject: Re: Neural Net cloak recognition
- Message-ID: <1992Nov23.043535.12879@organpipe.uug.arizona.edu>
- Date: 23 Nov 92 04:35:35 GMT
- Sender: nelson@soliton.physics.arizona.edu
- Organization: Lunar & Planetary Laboratory, Tucson AZ.
- Lines: 39
-
-
- >Wow, sounds great...first useful thing I've heard neural
- >nets used for. I would have otherwise replied by mail, but
- >you didn't have a mail address in your sig, and
- >I doubt <news@organpipe.uug.arizona.edu> would get to you...:)
-
- Dang, I should have anticipated that... my email address
- is nelson@soliton.physics.arizona.edu. There are quite a few
- interesting problems involving very high performance/speed
- pattern recognition in high energy physics, btw, if you are
- looking for something to fill your time. With the SSC coming
- out, these poor physicist are running around trying to find
- ways of determining which data is interesting and which is not
- (this decision needs to be done every 16 ns, of the 100MHz of
- events only about 10 Hz can be saved for consideration.)
-
-
- >So, which simulator did you use? Can I see the inputs?
-
- I used JETNET which is a ftp-able neural net simulator
- supposedly designed for use in high energy physics although
- I have come across nothing at all that limits it generality
- in the slightest. Hmm..... unfortunately looks like I lost
- the name of the ftp site, if anyone wants it e-mail me.
- The only thing real problem with JETNET is that it is written
- in FORTRAN (blech!).
-
- I'll mail you specifics directly, along with the program
- I used to generate the data. Looking back I see that there
- was one striking simplification that I should not have made,
- that is I assumed that the most recent update was always
- current (ie when the robot needs to spot a cloaker, it would
- always have just recieved an update). This allowed me to
- normalize the problem in terms of a reference point which
- the actual position *must* be within 1000 units of in either
- direction. This also offers an additional implied data point
- of 0.0,0.0,0.0. Maybe I'll try to rerun without this assumption.
-
- - Jeff (Miles Teg)
-