Where can I find examples of applications for trained ART and SOM networks?

Click For Summary

Discussion Overview

The discussion centers around the applications of trained Adaptive Resonance Theory (ART) and Self-Organizing Map (SOM) neural networks. Participants express a desire for examples of how to utilize these trained networks after the learning process, as well as seeking resources for further study on their applications.

Discussion Character

  • Exploratory
  • Technical explanation
  • Debate/contested
  • Homework-related

Main Points Raised

  • One participant expresses confusion about the next steps after training an ART or SOM network, seeking examples of practical applications.
  • Another participant suggests that the network should be trained with a specific application in mind.
  • A further reply elaborates on the potential uses of a trained network, mentioning clustering and classification as possible applications, and discusses the idea of creating internal representations of data.
  • There is a mention of the importance of identifying patterns through classification, which may aid in learning or in other applications.
  • One participant requests recommendations for books that cover applications of ART and SOM, indicating a desire to deepen their understanding of these concepts.

Areas of Agreement / Disagreement

Participants do not reach a consensus on specific applications or examples of trained ART and SOM networks. Multiple viewpoints on how to utilize the networks and the nature of their training remain present.

Contextual Notes

Participants express uncertainty regarding the specific types of data and applications they are working with, which may influence how the trained networks can be utilized. There is also a lack of clarity on the optimal representation of the models created by the networks.

Who May Find This Useful

Individuals interested in neural networks, particularly ART and SOM, and those seeking practical applications or further reading on these topics may find this discussion valuable.

eoghan
Messages
201
Reaction score
7
Hi!
I attended a class in neural networks and I learned something about ART and SOM.
I learned how to implement the learning algorithm, but then? I wasn't told how to use the trained net! I looked in Google and I didn't find much, everybody talks on how to train a neural net, but nobody says how to use it after the learning process.
Do you know a place where I can find examples of (simple) applications of trained ART and SOM?
 
Technology news on Phys.org
Er, it should be trained on its application, shouldn't it?
 
Yes, I was given a set of data and I was told how to train the net for those data. Now that I have a trained net for those data, what am I supposed to do? I mean, what can I do with that net?
 
eoghan said:
Yes, I was given a set of data and I was told how to train the net for those data. Now that I have a trained net for those data, what am I supposed to do? I mean, what can I do with that net?

What kind of data are you working with?

Depending on the data and the application, the idea will be to create a neural network that is able to to create some kind of internal representation of a system that should converge to the represent the actual output itself.

If you are trying to create for example a clustering or classification model, then what happens is that the initial data will 'train' the network to create a model for taking an input and classifying either in an unsupervised form (clustering) or in a supervised form (classification) so that with that model, you can feed in some input and the network will classify it according to how it was trained with the training data.

The classification of data (unsupervised or supervised) helps not only for learning but for the identification of patterns. By creating the right classifications (supervised or unsupervised) you are finding patterns and you can use this for learning either in the network, or in another application (like taking the results of the network and using it in another application).

If the neural network is an optimal representation of the model that you are analyzing, then what this network will represent is the minimal representation of the model in it's most compressed form. Most likely the model will not be an errorless 1-1 representation of the model (there will be likely be errors of some sort), but again the point is that the network itself is a good approximation.

How you actually make use of this depends on exactly what you had in mind when you trained the data and created the network and what the actual data is in context.
 
Uhm... I think I have a deep lack of knowledge of som and art... do you know any good book with applications where I can study them?
 

Similar threads

  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 18 ·
Replies
18
Views
4K
  • · Replies 4 ·
Replies
4
Views
1K
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K
  • · Replies 7 ·
Replies
7
Views
8K
  • · Replies 3 ·
Replies
3
Views
6K
  • · Replies 3 ·
Replies
3
Views
2K
  • · Replies 1 ·
Replies
1
Views
2K