Training Neural Networks using Multi-Class output

The Deep Learning toolbox supports classification based training (from feature based data) for ony 1 label per sample. I have a MxD training set (D number of features and M number of samples). Each output should be characterized by 'T' number of labels (ie final output MxT). My question is how do i get around this limitation ? (The labels are mutually exclusive)

Answer 1

[Raynier Suresh]

One way to obtain multiple labels for a single sample is to branch the network and have multiple classification layers or regression layers.

Refer the below link for designing and training multi-output networks :

https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html#responsive_offcanvas

[Anirudh Roy]

Branching isnt probably ging to solve my problem, stacking of classification layer would. But yeah i get the idea, ill have to customize the whole thing (loop and layer)