Build deep learning model for non-image data

Question

Can you please tell me how to build an autoencoder using CNN and pooling layers, with a single matrix(4,4) with integer numbers? e.g, input data = array([[ 4, 3, 8, 6], [ 1, 1, 2, 2], [24, 18, 32, 24], [ 6, 6, 8, 8]])
autoencoder(data)
output data= array([[ 4, 3, 8, 6], [ 1, 1, 2, 2], [24, 18, 32, 24], [ 6, 6, 8, 8]])
Explanation: https://medium.com/machine-learning-researcher/auto-encoder-d942a29c9807

Answer 1

The article you quoted is already well detailed, all you need to do is replaced linear units by convolution and max pool in the encoding part and deconvolution and upsampling in the decoding part.

Here is in example using Keras of a convolutional autoencoder.

But if you're new to Deep Learning I would not advise you to start here. Autoencoder are tricky, the expected output you described above could easily be obtained by setting:

def autoencoder(x):
    return x

Which is not something you want.

People commonly design "undercomplete" autoencoders, with internal layer of smaller dimension to induce some compression. But "overcomplete" autoencoders with larger internal dimension are also valid architecture combined with regularization.