Performance of linear model using embedded features

[macaodha]

Hi there,

I really like the paper and was wondering if you tried training a linear SVM on your 128 dim feature space - and if so how will did it perform compared to the standard softmax approach.

Thanks