Multi-label classifications via deep learning tool box

I'm working on a project where I am going to estimate DOAs of mutiple signals via CNN, and I modeled it as a multilabel classification task, which means that each label or category reveals whether there is a signal arriving from one of the many potential directions. Since there could be multiple signals, the response of the network should also be multiple categories. The real problem is function trainNetwork wants the responses to be in a single column: one category per image(in my case it is a covariance matrix instead). What can I do with the function trainNetwork so that the network could be trained to give out mutilple categories?

Answer 1

[Hiro Yoshino]

why don't you consider this problem to be regression problem since the one you are trying to estimate is a covariance matrix, which is obviously a set of continous nubers.

Input is an image-like thing and the output is, therefore, the associated covariance matrix.