How Can We Determine - PowerPoint PPT Presentation

1 / 30
About This Presentation
Title:

How Can We Determine

Description:

Nelder Mead Simplex Algorithm: The crawling spider. high. high. high. high. Simplex difficulties: Huge use of storage (n2) Equivalent weight to all variables ... – PowerPoint PPT presentation

Number of Views:36
Avg rating:3.0/5.0
Slides: 31
Provided by: alankayl
Learn more at: http://www.cs.utexas.edu
Category:

less

Transcript and Presenter's Notes

Title: How Can We Determine


1
How Can We Determine Numerical Quality in
Software?
The Doctoral Dissertation of Thomas Rowan
2
Thomas Harvey Rowan Functional Stability
Analysis Of Numerical Algorithms Doctoral
Dissertation Department of Computer Sciences The
University of Texas at Austin 1990
3
What is meant by numerical quality?
4
What is meant by numerical quality?
The computed solution (i.e. the output) is the
exact solution to a close problem (i.e. having
slightly different input)
5
What is meant by numerical quality?
The computed solution (i.e. the output) is the
exact solution to a close problem (i.e. having
slightly different input)
This is called numerical stability
6
Input a pair of lines in the plane
7
Input a pair of lines in the plane
Output the intersection (if any)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
This is what we seek A poor solution to a well
conditioned problem
15
?
?
Notice the difficulty We have to guarantee
that the approximate solution is not the
solution to ANY close problem
16
Idea Search the input space attempting to
maximize the ratio of the forward error to the
condition number.
17
Idea Search the input space attempting to
maximize the ratio of the forward error to the
condition number.
Details, details How do you get the forward
error? How do you get the condition number?
18
Would you like to find the maximum of this?
19
Nelder Mead Simplex Algorithm The crawling
spider
20
high
high
high
high
21
(No Transcript)
22
(No Transcript)
23
  • Simplex difficulties
  • Huge use of storage (n2)
  • Equivalent weight to all variables

24
  • Simplex difficulties
  • Huge use of storage (n2)
  • Equivalent weight to all variables

Subplex Determine small subspaces likely to be
most useful for determining minimum
25
The idea is to use the Subplex Algorithm
to locate minima of (the negative of) the
backward error
SUBPLEX
INSTAB
code
Backward Error
code extra precision
26
Does it work?
27
Does it work?
  • Tried on known unstable algorithms
  • Tried on known stable algorithms
  • Tried on unknown algorithms

28
Does it work?
  • Tried on known unstable algorithms

Identified all as unstable
  • Tried on known stable algorithms

Found no false positives
  • Tried on unknown algorithms

Discovered unstable algorithms
29
Backword error estimate for Gaussian elimination
30
Anything left to do?
  • Two papers to write
  • Translation into C or Matlab
  • More experiments
  • Refinements to INSTAB or SUBPLEX
Write a Comment
User Comments (0)
About PowerShow.com