[Webinar] Streamline your web hosting managementRegister Today

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 361
  • Last Modified:

In Neural Network training, how to represent instances that do not belong to any class

Also, how should that be reflected into the design of the network itself?
I mean if the output of 0001 means 1 and the output of 0010 means 2, can we train the network to produce 0000 for instances that do not belong to any class?
0
Lamiaaa
Asked:
Lamiaaa
  • 2
2 Solutions
 
GwynforWebCommented:
Are you using a forward-feeding (ie layers proceeding left-right) or a feedback network?
0
 
mccarlIT Business Systems Analyst / Software DeveloperCommented:
Theoretically, yes, 0000 would mean no class, but in practical terms, you would have to evaluate the performance of that. It would depend on all the usual things with NN's, such as the input data, number of layers, number of neurons, how you train it, etc.
0
 
LamiaaaAuthor Commented:
Thanks guys,

yes, it is forward-feeding.

Are there any sampling approaches (proposed to solve similar problems) to build a training set that appropriately represent the 0000 cases?

It seems difficult because "every thing else" is a very huge and diverse set!
0
 
mccarlIT Business Systems Analyst / Software DeveloperCommented:
It is a bit hard to say, as only you know your problem domain. One possibility might be a random sample that is the same size as the total of all 'classified' samples, ie. of the total training set, half is "everything else' and the other half is 'one of the target classes'.

I would base this on the performance of the network as it stands. So first, you would include these 'everything else' inputs into your verification set, and see how the network performs. Then depending on the results of the verification step, modify the training set to see what improvements can be gained.
0

Featured Post

Free Tool: Port Scanner

Check which ports are open to the outside world. Helps make sure that your firewall rules are working as intended.

One of a set of tools we are providing to everyone as a way of saying thank you for being a part of the community.

  • 2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now