Best Automated Method for Selecting Tikhonov Regularization Parameter?

  • Context: Graduate 
  • Thread starter Thread starter newbee
  • Start date Start date
  • Tags Tags
    Regularization
Click For Summary

Discussion Overview

The discussion revolves around the selection of the regularization parameter in Tikhonov regularization, focusing on automated methods and the underlying principles that inform parameter choice. It touches on theoretical aspects as well as practical applications in modeling.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested

Main Points Raised

  • One participant questions the existence of a general method for selecting the regularization parameter, suggesting that it often relies on physical understanding of the modeled system.
  • Another participant notes that in the context of inverting the heat equation, increasing the regularization parameter can smooth out high-frequency solutions but may also obscure actual high-frequency information.
  • A different participant mentions that there is extensive research on this topic within the field of machine learning.
  • A request for references to the work on automated selection methods is made by another participant.

Areas of Agreement / Disagreement

Participants express differing views on the existence and nature of automated methods for selecting the regularization parameter, indicating that multiple competing perspectives remain unresolved.

Contextual Notes

The discussion highlights the dependence on physical understanding and the balance between smoothing and information recovery, but does not resolve the mathematical or practical implications of these factors.

newbee
Messages
80
Reaction score
0
What is the best automated way to select the regularization parameter in a Tikhonov regularization?
Can you point me toward some code for this purpose?

Thank you,
 
Physics news on Phys.org
I don't think there is a general method for determining the parameter. Usually the number you choose comes from a physical understanding of the system being modeled.

In the classic example of inverting the heat equation, increasing the regularization parameter smooths out spurious high frequency solutions, but at the same time prevents you from recovering any actual high frequency information that might have been present to start with. The regularization parameter is chosen to properly to balance these competing factors.
 
Last edited:
There's a huge amount of work on exactly this question in the field of machine learning.
 
Cincinnatus

Do you have any references to the work? Much appreciated!
 

Similar threads

Replies
6
Views
1K
  • · Replies 5 ·
Replies
5
Views
3K
Replies
4
Views
1K
  • · Replies 5 ·
Replies
5
Views
3K
  • · Replies 6 ·
Replies
6
Views
3K
  • · Replies 5 ·
Replies
5
Views
1K
  • · Replies 7 ·
Replies
7
Views
3K
  • · Replies 2 ·
Replies
2
Views
3K
  • · Replies 4 ·
Replies
4
Views
2K
  • · Replies 7 ·
Replies
7
Views
4K