Multi-objective recursive least squares

  • Context: Graduate 
  • Thread starter Thread starter Superfish
  • Start date Start date
  • Tags Tags
    Least squares Squares
Click For Summary
SUMMARY

The discussion centers on the implementation of a multi-objective recursive least squares solution using the Woodbury matrix identity. The original poster encounters challenges when applying the identity for rank 2 updates, prompting inquiries about existing derivations. A generalization of Woodbury's identity is provided, which allows for rank m updates, emphasizing the necessity of inverting a rank m matrix. The conversation highlights the importance of understanding matrix inversion techniques in recursive least squares applications.

PREREQUISITES
  • Understanding of recursive least squares algorithms
  • Familiarity with the Woodbury matrix identity
  • Knowledge of matrix inversion techniques
  • Basic concepts of multi-objective optimization
NEXT STEPS
  • Study the derivations of the Woodbury matrix identity for rank m updates
  • Explore applications of the matrix inversion lemma in optimization problems
  • Learn about recursive algorithms in statistical modeling
  • Investigate multi-objective optimization techniques in machine learning
USEFUL FOR

Researchers, data scientists, and mathematicians involved in optimization problems, particularly those working with recursive algorithms and matrix computations.

Superfish
Messages
1
Reaction score
0
Is this possible?

I've computed a multiobjective least squares solution and want to make it able to be updated recursively but I get stuck at applying the woodbury matrix identity since it's no longer a rank 1 udpate. Are there any derivations of this anywhere or is this not possible? Thanks
 
Physics news on Phys.org
Of course, if you have a rank 2 update, you can always use Woodbury's twice in a row, etc.

A generalization of Woodbury's identity is
[tex] (A+BCD)^{-1}=A^{-1} - A^{-1} B(D A^{-1} B + C^{-1} )^{-1} D A^{-1}[/tex]
Where A is nxn, B is nxm, C is mxm, and D is mxn, so this can work for a rank m update. It isn't too suprising that you have to invert a rank m matrix to do it.


jason

EDIT: just realized that Woodbury's was never stated. In the above, if B is nx1 vector, D is the transpose of B and C is just the number 1, then you have Woodbury's identity. The generalization above is usually called the matrix inversion lemma
 
Last edited:

Similar threads

  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 2 ·
Replies
2
Views
2K
Replies
8
Views
4K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 4 ·
Replies
4
Views
4K
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 30 ·
2
Replies
30
Views
3K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 7 ·
Replies
7
Views
4K
  • · Replies 8 ·
Replies
8
Views
2K