Backpropogation in Neural Network Toolbox

I am trying to implement an Autoencoder to recognize MNIST images. I don't want to use autoencoder available in Neural Network toolbox but just want to use the backpropogation method. How can I use backpropogation method implemented in Neural Network Toolbox in my own created Autoencoder?

Answer 1

[Gautham Sholingar]

Hi Neelabh,

It is important to realize that the back-propagation methods in the neural network toolbox operate on network classes which are specific to MATLAB and it would be inconvenient to modify your model to ensure that it is compatible with these methods. There are a few ways of solving your problem:

1) Backpropagation for MNIST image classification is a well-studied subject and you should be able to find lots of examples online explaining this in detail along with custom code snippets.

2) The other approach is to use MATLAB Auto-encoders which have been applied to the exact problem you are trying to solve as shown in the following example:

https://www.mathworks.com/help/nnet/examples/training-a-deep-neural-network-for-digit-classification.html

I hope that this is a good starting point for you to think about this problem.

Regards,

Gautham Sholingar

[Neelabh Pant]

Hi Gautham, thank you for answering the question. Like you mentioned, I tried working on trainAutoencoder class in Neural Network Toolbox but faced another challenge. I posted the question related to that here,

http://www.mathworks.com/matlabcentral/answers/314392-autoencoder-encoderweights-and-decoderweights

Do you think you know the answer to that. Thanks again!