From: n.o.s.p.a.m.-mvcs@idt.net (MVCS)
Subject: Re: Mahalanobis Metric and ANNs.
Date: Mon, 22 Nov 1999 04:36:16 GMT
Newsgroups: sci.math.num-analysis

almada_negreiros_2@my-deja.com wrote:
>...
>Do you know some previous work on the implementation of
>the Mahalanobis Metric with Artificial Neural Networks?
>...

I use Mahalanobis metrics to estimate linearity of data swarms. This
assists determination of how non-linear a problem might be, and helps
to interpret linear regression (using SVD) compared to NN trianing
results. For most problem sets I've encountered since I instituted
Mahalanobis stats, I've found that SVD linear results are just about
as good as NN results. And as a pragmatic engineering reality, the
ability to relate to and understand the linear results outwieghs any
VERY slight advantage of NNs (again, for the problem sets in the
industry in which I work).