I am trying to use a simple perceptron to recognize if there is a square or a circle on an image. The images I generated are 300x300 px and I am having issues training the network since the images are large and I seem to run out of memory after a few images.

What would be the correct approach for this?

enter image description here enter image description here

  • 1
    $\begingroup$ Please edit the question to provide more context. 1. What are the constraints that we must work within? The correct approach would be to not use a neural network at all, and instead use a dedicated method. Is that acceptable? If not, what exactly are the limits on what is acceptable? 2. What exactly are you doing now? What is the architecture of your perceptron? One layer? Fully connected? What are the inputs? 90,000 inputs, one per pixel? What are you using to train the network? (continued) $\endgroup$ – D.W. Nov 23 '16 at 18:15
  • 1
    $\begingroup$ Are you using stochastic gradient descent (SGD)? Why are you running out of memory? SGD should use memory that is a proportional to the number of weights/parameters in the model, regardless of number of training images. This sounds more like an issue related to the particular implementation or software you're using (which is off-topic here), rather than anything related to science or concepts. It should be easy to train such a network if you use proper methods and software. 3. What other approaches have you considered, and why have you rejected them? $\endgroup$ – D.W. Nov 23 '16 at 18:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.