LSTM with multiple Softmax layers

Hi,

I am working with LSTM model. It receives sequences of N users with B features [matrix of N*B ]. And I would like to generate outputs in form of sequences with N users and 3 labels [matrix of N*3]. Indeed, I would like to perform 3 different classification : 3 multi-class of labels

The implementation here allows me to have output sequences in the form of 1 vector [matrix of N*1]. I guess it is because I am using only one softmax layer. Is there any way to work with 3 softmax layers in the output or any other solution to generate 3 multi-class of labels ?

layers = [ ...

sequenceInputLayer(numFeatures)

bilstmLayer(numHiddenUnits,'OutputMode','sequence')

fullyConnectedLayer(numClasses)

softmaxLayer

classificationLayer];

Answer 1

[Bhargavi Maganuru]

Hi,

If you need multi-class label, you can specify numClasses and include a fully connected layer of size numClasses. As the last layer is a ClassificationLayer, the ouput will be in the form Nx1 vector, where each value represents the class to which it belongs.