A multi-layered support vectors machine

My work on Tempus?during?the past 5 years included developing a multi-layered SVM by using a support vector regressor, acting as the kernel of another SVM. The support vector regression machine that is used as a kernel, is trained on a dataset produced from the ideal kernel matrix for the given problem. This dataset is called a?support vectors manifold. An incorrect example implementation of a manifold dataset can be seen in?manifolds.cpp. The ideal kernel matrix is generated from the difference of labels and by concatenating the feature vectors, or?L1 - L2 -> F1 § F2, as you can see in OnlineMIMOSVR::get_reference_kernel_matrix(). But this doesn't work as I assumed, so I?will?need?help here. Anyone that could chime in with useful comments is welcome.

Another approach that Prof. Emanouil Atanasov provisioned using Google's feed-forward implementation of ANN in TensorFlow to approximate an ideal matrix, which is implemented like this.