I wish to design my own autoencoder with some input arrays I specify. let us say, I have 10,000 2D arrays in which elements are 1's and 0's stored in a folder. Because the autoencoder object in MATLAB contains only two hidden layers, yet I am going to contain loads of other types of layers.
As I have seen in the doc, if I use ImageDatastore object, I will have to label each folder in that images are pre-classified to, yet as we know, an autoencoder aims to replicate the input and extract useful features in the hidden layers rather than classifying them. So my first question is how to input my arrays sequences?
My second question is that if I use the trainNetwork(...) function, again it requires pre-labelled images and folders, hence if my input arrays are for my autoencoder only, I surmise that either costs or backpropagated gradients will be calculated properly. So what should I do?
It will be equally great that if someone could provide me with an example or some texts in the doc so that I will be able to figure them out myself, if a full answer is too tiresome for you.