I am trying to apply Probabilistic Neural Network (PNN) on MNIST dataset using MATLAB on 28000 samples. I tried to build the model but it gave an error saying matrix dimension exceeds memory showing it will require 5.08 GB (28000x28000) matrix.

I want to know should I decrease the number of samples as input for training from 28000 to 200 samples as I read in research paper.

Should I reduce the sample size or is there something else wrong?

is pnn only applicable on small data sets?

  • $\begingroup$ No ,to what size I should reduce? $\endgroup$
    – Boris
    Dec 19, 2017 at 14:10
  • $\begingroup$ I will reduce size to 200 samples and vector size is 784 $\endgroup$
    – Boris
    Dec 19, 2017 at 14:13
  • $\begingroup$ Is it possible to do batch training, where you repeatedly feed in training sets with relatively few samples? $\endgroup$ Dec 19, 2017 at 14:18
  • $\begingroup$ @NuclearWang maybe not tried it yet $\endgroup$
    – Boris
    Dec 19, 2017 at 14:19

1 Answer 1


Reducing the sample size is pretty much up to you. PNN have the drawback that the size of the network after training is equal to the size of the training dataset. Therefore it's a bit impractical to use for large datasets. I also believe that there is evidence that suggest the performance may also decrease when the layers are big.

I don't have access to the paper but I imagine a big reason why they reduced the sample was to avoid the memory error that you're encountering right now. I also suspect they carefully selected a good representation for the sample.

  • $\begingroup$ I added a paper in the question.they use 200 samples $\endgroup$
    – Boris
    Dec 19, 2017 at 14:31
  • $\begingroup$ I should have clarified. I see the link but the issue is that it requires payment to download. I don't intend to spend 30 dollars to read a paper from 11 years ago. $\endgroup$
    – Tophat
    Dec 19, 2017 at 14:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.