🚀 go-pugleaf

Many readers of the list may have retrieved and read Brian Ripley's
fascinating chapter "Statistical aspects of neural networks".  This
chapter provided a statical analysis of neural network classification
and pattern recognition techniques.  (Sorry, I no longer have an
online copy or information about the ftp site.)  In brief, Ripley's
analysis suggested that neural computing models don't really get you
anyplace you can't go using sophisticated statistical methods for
classification and pattern analysis.

What I want to know is: have any neural computer people written a
rejoinder to Ripley?  It's a huge subculture / industry, one would
think someone would rise to defend the field.

                         *   *   *

Since you were kind enough to read this far, you might be interested
in this reference --

Weinstein, J., et al. (1992). Neural computing in cancer drug
  development: predicting mechanisms of action. _Science_, _258_,
  447-451.

This article compares the predictive performance of a neural network
to a linear discriminant model, and finds the former superior (which
may only mean that the problem is nonlinear).


 [][][][][][][][][][][][]   William Gardner   [][][][][][][][][][][][][][][]
 []   /_   o / /  Psychiatry Dept, School of Medicine      412-681-1102   []
 []  /__) / / /       University of Pittsburgh   wpg@ethics.med.pitt.edu  []
 [][][][][][][][][][]    Pittsburgh, PA 15213  [][][]  FAX:412-624-0901 [][]
-- 
 [][][][][][][][][][][][]   William Gardner   [][][][][][][][][][][][][][][]
 []   /_   o / /  Psychiatry Dept, School of Medicine      412-681-1102   []
 []  /__) / / /       University of Pittsburgh   wpg@ethics.med.pitt.edu  []
 [][][][][][][][][][]    Pittsburgh, PA 15213  [][][]  FAX:412-624-0901 [][]