I have read many articles which replaced cross-entropy or softmax layer with SVM classifier in neural network. Their results suggested that NN-SVM (SVM as classifier) outperformed softmax, cross-entropy.
(One such paper - http://deeplearning.net/wp-content/uploads/2013/03/dlsvm.pdf)
I wish to train and test NN-SVM on my datasets. I have two questions
- Is there any existing library/code available
- Can some advice on how to implement this. Later, I can make it public for others also