Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!agate!news.ucdavis.edu!sunnyboy.water.ca.gov!sunnyboy!nsandhu
From: nsandhu@venice.water.ca.gov (Nicky Sandhu)
Subject: Determining range of output with a certain probability
Message-ID: <NSANDHU.94Nov21120700@grizzly.water.ca.gov>
Sender: news@sunnyboy.water.ca.gov
Organization: Calif. Dept. of Water Resources
Date: Mon, 21 Nov 1994 20:07:00 GMT
Lines: 28

Hi,
     
     I am trying to simulate a system using a neural network. The
inputs and outputs of the network are real valued time series. I want
to be able to specify the range within which the output will lie with a
certain probability P. I would like your comments on the
following method.:-

1. Randomly divide the data set into k- segments.

2. Train on one of them to covergence. (or using a cross-
   validation set)

3. Obtain the errors from simulating the trained network on the
   other data segments.  

4. Estimate the distribution of those errors.

5. Use that distribution to come up with a range of values within
   which the output will lie with probability P.

6. Repeat steps 2-5 for different sets for training.

Is there any other more efficient way of computing such a range
of values? A more statistically justifiable method?? Any
references are welcome!!

-Nicky
