Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!gatech!concert!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: MLP vs Projection Pursuit
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <D0Cqvy.LEC@unx.sas.com>
Date: Mon, 5 Dec 1994 19:16:46 GMT
References:  <66979.alison@atp.biochem.usyd.edu.au>
Nntp-Posting-Host: hotellng.unx.sas.com
Organization: SAS Institute Inc.
Lines: 33


In article <66979.alison@atp.biochem.usyd.edu.au>,  <alison@atp.biochem.usyd.edu.au> writes:
|> Given a nonparametric regression problem in which there are multiple,
|> continuous-valued inputs and outputs (i.e., multivariate / multiresponse)
|> what are the pros and cons of using a multilayer perceptron (with sigmoid
|> hidden units and linear output units) and projection-pursuit regression
|> (with multuple outputs).

The obvious advantage of an MLP is that it is easier to compute
predicted values (i.e. outputs). I would speculate that projection
pursuit regression is less prone to local minima. However, I know of
no studies that have performed nontrivial comparisons of these two
methods.

|> Are there other nonparametric statistical methods that can be used for this
|> regression problem?

Numerous ones. There are many varieties of fully nonparametric
regression such as kernel and k-nearest-neighbor methods. There are
local polynomial methods such as LOESS. There are other types of
semiparametric models such as generalized additive models. There are
flexible parametric models such as tensor splines. There are tree-
based methods such as CART and MARS. One handy reference is:

   Haerdle, W. (1990), _Applied Nonparametric Regression_, Cambridge
   Univ. Press.


-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
