"DeepTree" network with binary classifiers at each node

Question

I need to do a project where I build a "DeepTree Network". Let's say I have a 3-classification problem with classes A, B, and C.

My tree should then look like this:

• first node: CNN binary classifier classifying between class C, and superclass A+B

• second node: CNN binary classifier classifying between class A and B

Does anyone know if there are any existing papers on this subject? Or any implementation? The leaves don't need to be CNNs necessarily, any sort of tree implementation like this in TensorFlow would help me a lot.

Answer 1

You can try the SVM with your decision function :

https://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html#sklearn.svm.SVC

it can handle muilt classfication problem with default ovr decision function

if you stick with tensorflow you can only try linver svc : https://www.tensorflow.org/api_docs/python/tf/contrib/learn/SVM

Answer 2

This seems sufficiently specialized that it's unlikely to have an existing implementation. It should be easy to do by hand. If you really needed to, it might be possible to frame it as a stacked ensemble that could fit into an existing implementation.