Path: utzoo!yunexus!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!uakari.primate.wisc.edu!ames!pacbell.com!pacbell!srv.PacBell.COM!sire
From: sire@srv.PacBell.COM (Sheldon Rothenburg)
Newsgroups: comp.ai
Subject: Re: Neural Nets that are WORKING
Summary: Re: Baggage Checking
Keywords: neural network
Message-ID: <649@pbhal.PacBell.COM>
Date: 29 May 90 18:15:24 GMT
Article-I.D.: pbhal.649
References: <25089@netnews.upenn.edu> <1990May21.165517.29866@hri.com>
Organization: Pacific * Bell, San Ramon, CA
Lines: 37

In article <1990May21.165517.29866@hri.com>, rolandi@sparc9.hri.com (Walter Rolandi) writes:
> 
> How many neural nets are actually in production, regardless of application,
> either in medicine, finance, manufacturing, ---anything?
> Are folks really applying neural nets in any profitable way?

I didn't save the poster's comment, but someone referred to an SAIC 
developed, neural network based baggage checker as a successful application
of neural nets. It does use neural nets to analyze the results
of gamma ray bombardment of the air traveller's luggage. It looks for
evidence of minute quantities of "escaping" nitrogen which fit the 
profile of plastic explosives.

The NY Times discussed the commercial viability of this system in a
recent business technology section. It was mandated last year for
purchase and installation at 40 US Airports by the FAA. There has been
no compliance. The airlines and the airports do not want to foot the bill
for these van-sized, $1M units. The units would require passengers to
wait longer and airports are frequently short of space. There is reportedly
also the issue of an unacceptable number of false positives. The most
egregious common error was mistaking the nitrogen release from women's
high heel shoes for plastic explosives. This has allegedly been rectified.

Six units were purchased by the US Government. They are the only purchasers
so far. The machines are in use in London's Heathrow, New York's Kennedy,
and one other airport (I think it was Miami's).

So---this is not an unqualified success. My understanding is that much of
the successes are in classification types of domains (signal detection)
with the kind of military application that would not be widely
publicized. Note who funds this research, largely DARPA, ONR etc.

Whew! This was longer than I expected. I hope it was helpful.

Shelley

te