Automatic pattern recognition and data modeling

Click For Summary

Discussion Overview

The discussion revolves around the concept of automatic pattern recognition and data modeling by computers, exploring potential applications in various fields such as forecasting and data analysis. Participants examine existing technologies and the challenges of accurately identifying patterns in data sets.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested

Main Points Raised

  • Some participants propose that future advancements may enable computers to recognize patterns in data and fit equations to those patterns, with applications in fields like stock forecasting and air traffic control.
  • Others mention existing technologies, such as YouTube's ability to recognize audio and video patterns for copyright purposes, as a form of pattern recognition.
  • A participant suggests that mathematical tools like MATLAB may already provide data fitting and smoothing models to suggest appropriate equations for data sets.
  • Another participant argues that identifying a linear relationship from a small data set is ambiguous without additional information, noting that a cubic equation may fit better than a linear one for the provided data.
  • Lowess curve fitting is introduced as a method that uses multiple equations for curve fitting rather than a single equation, emphasizing a more flexible approach to modeling data.

Areas of Agreement / Disagreement

Participants express differing views on the capabilities of current technologies and the feasibility of automatic pattern recognition. There is no consensus on the best approach to modeling data or the limitations of existing methods.

Contextual Notes

Participants highlight the importance of additional information in model selection and the limitations of small data sets in determining relationships, indicating that assumptions and constraints play a critical role in data modeling.

moonman239
Messages
276
Reaction score
0
I can't help but think that some day, someone will figure out a way to get a computer to recognize patterns in a given set of data, and fit an equation, if there is any, to that set.

Such a system could be used in areas like forecasting stocks, weather, and sales, or even in air traffic control, which would be by far the coolest use of such a system.
 
Technology news on Phys.org
youtube can already "recognize" some video or audio patterns to search for "matched 3rd party content" in submitted videos, mostly used to include ad's on videos where some of the profit for the ad goes to the "3rd party content" copyright owner.
 
rcgldr said:
youtube can already "recognize" some video or audio patterns to search for "matched 3rd party content" in submitted videos, mostly used to include ad's on videos where some of the profit for the ad goes to the "3rd party content" copyright owner.

I see. However, I'm talking about having a computer look at a data set such as:

2,4.2,6,8.5

and figure out that the data can be best modeled by a linear equation.
 
moonman239 said:
I see. However, I'm talking about having a computer look at a data set such as: 2,4.2,6,8.5 and figure out that the data can be best modeled by a linear equation.
Doesn't MATLAB and/or some other mathematical tools have a data fitting / smoothing model that at least suggests what type of equation would be best to do this?
 
moonman239 said:
I see. However, I'm talking about having a computer look at a data set such as:

2,4.2,6,8.5

and figure out that the data can be best modeled by a linear equation.

A computer can't do that for precisely the same reason that a human can't: it's impossible to unambiguously identify that as evidence of a linear relationship unless you supply more information to constrain the model selection. In fact, if you try to fit functions to that particular set of data you'll find that a cubic equation is a better fit than a linear equation.

As to your original question, lots has already been done: artificial neural nets, support vector machines, kernel methods more generally, and many others are all in widespread use in pattern recognition.
 
There is Lowess curve fitting that with cross-validation looks at data and does a curve fit by gluing together whatever fits the best range-wise, in other words, it does not use a single equation, but several for a best fit everywhere.
 

Similar threads

  • · Replies 7 ·
Replies
7
Views
3K
Replies
1
Views
2K
  • · Replies 3 ·
Replies
3
Views
5K
  • · Replies 4 ·
Replies
4
Views
2K
  • · Replies 3 ·
Replies
3
Views
2K
Replies
1
Views
2K
Replies
29
Views
5K
  • · Replies 12 ·
Replies
12
Views
4K
  • · Replies 4 ·
Replies
4
Views
2K
  • · Replies 36 ·
2
Replies
36
Views
5K