Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!newsfeed.internetmci.com!in2.uu.net!bcstec!nntp
From: Mike Fouche <mikef@hsvaic.hv.boeing.com>
Subject: Re: Q: NN in safety critical applications ?
X-Nntp-Posting-Host: e677744.hv.boeing.com
Content-Type: text/plain; charset=us-ascii
Message-ID: <DMpsnr.Mrv@bcstec.ca.boeing.com>
Sender: nntp@bcstec.ca.boeing.com (NNTP News Access)
Content-Transfer-Encoding: 7bit
Organization: Missiles & Space Division
References: <4fnr16$pss@wmwap1.math.uni-wuppertal.de>
Mime-Version: 1.0
Date: Tue, 13 Feb 1996 12:43:02 GMT
X-Mailer: Mozilla 1.2 (Windows; U; 16bit)
Lines: 26

You're right - there is no 100% reliability.  In terms of aircraft, most accidents
are caused by the pilot anyway, regardless of the guidance, navigation & control 
sytem.

Still, this is one of the tough problems with neural nets - people want 
"deterministic" proof of stability.  What actually happens with "deterministic" 
systems is that the system is tested in all possible modes.  Doesn't matter if it's 
a satellite control system or missile guidance & control system - what is typically 
done is a rigorous schedule of testing.  I've seen faulty control laws designed - 
they were determined to maintain stability to the system, mind you, - however the 
designer failed to take into account other factors.  

We've been applying neural networks to control & guidance for about 6 years now.  
From what I've seen, neural nets are very stable in these applications.  In fact
the performance tends to drop off more more gradually than conventional solutions.

So instead of focusing on new techniques, the traditional methods of testing, such
as Monte-Carlo simulations and Covariance analysis, should be emphasized.

That's my 2 cents worth.

Mike Fouche
Boeing Missiles & Space Division



