Optimizing Across Noisy Domain

  • Context: Graduate 
  • Thread starter Thread starter tangodirt
  • Start date Start date
  • Tags Tags
    Domain
Click For Summary

Discussion Overview

The discussion focuses on methods for optimizing a function across a noisy 2D surface. Participants explore various approaches and considerations for numerical optimization without fully computing the surface or filtering it.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested

Main Points Raised

  • One participant inquires about established methods for optimizing across a noisy 2D surface without computing the entire surface or filtering it.
  • Another participant suggests using a median filter as a potential technique for filtering noise.
  • Multiple participants emphasize that the definition of 'optimize' is ambiguous and depends on factors such as the criterion for optimality, the form of the expression being optimized, and specific constraints or requirements of the problem.
  • A participant notes that if the function is noisy, the observed maximum or minimum may not reflect the true optimal value, raising questions about the characteristics of the noise involved.
  • One participant proposes the use of Singular Value Decomposition (SVD) but acknowledges that it may still require computing the entire surface.

Areas of Agreement / Disagreement

Participants generally agree that the term 'optimize' requires clarification and that the characteristics of the noise and the specific requirements of the optimization problem are critical to determining an appropriate approach. However, there is no consensus on a specific method or solution.

Contextual Notes

Participants highlight the need for more details regarding the optimization criteria, the nature of the noise, and the constraints of the problem to provide more tailored suggestions.

Who May Find This Useful

Individuals interested in numerical optimization, particularly in contexts involving noisy data or surfaces, may find this discussion relevant.

tangodirt
Messages
51
Reaction score
1
Are there any established methods for optimizing across a 2D surface with noise? I am trying to find the maximum across a 2D surface, but the surface is extremely noisy. Ideally, I would numerically optimize a function without resorting to computing the entire surface, filtering the surface, and searching for a maximum, but I am not finding any established methods for this.

Any ideas?
 
Mathematics news on Phys.org
'Optimize' can mean almost anything. The right thing to do will strongly depend on

1. The criterion by which you are defining optimality ,

2. The form of the expression you are optimizing, and

3. Any other detauls that matter: constraints, continuous or discrete space, whether this is somethig that must be solved many time very quickly or if it just done once in awhile and can run a long time to converge, etc.

Unless you provide more details folks here cannot do much to help you.

Jason
 
jasonRF said:
'Optimize' can mean almost anything. The right thing to do will strongly depend on

1. The criterion by which you are defining optimality ,

2. The form of the expression you are optimizing, and

3. Any other detauls that matter: constraints, continuous or discrete space, whether this is somethig that must be solved many time very quickly or if it just done once in awhile and can run a long time to converge, etc.

Unless you provide more details folks here cannot do much to help you.

Jason

1. Maximize/minimize over a known domain.

2. It is a generic function. A black box with two inputs that returns an output that is noisy.

3. Something that needs to be solved many times, very quickly.

Think of this as Excel's "Solver", but with a noisy function.
 
I've never used excel's solver - you aren't describing what 'optimal' means. If your function is noisy, the maximum or minimum will likely be due to noise, not what you care about. So ... what does 'optimal' mean in this instance? What do you know about the problem (characteristics of noise, etc.)?

jason
 
Would the SVD be helpful? It would presumably still require you to compute the entire surface, though.
 

Similar threads

  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 5 ·
Replies
5
Views
2K
  • · Replies 6 ·
Replies
6
Views
3K
  • · Replies 12 ·
Replies
12
Views
2K
  • · Replies 24 ·
Replies
24
Views
2K
  • · Replies 1 ·
Replies
1
Views
508
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 5 ·
Replies
5
Views
415
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 18 ·
Replies
18
Views
2K