Suppose we have a multi-class classification problem, where the number of classes $K \geq 3$

We use a tree structure of multiple SVMs to divide and conquer the problem, with one example in the figure below.

example tree structure

Generally, in this kind of hierarchical classification, if a sample is wrongly classified in an upper layer, it is less likely to be corrected again in lower layers.

My question: Is there any strategy to mitigate such hierarchical error propagation in tree-structured classification?

Much appreciated!

  • $\begingroup$ Interesting question. Have you tried weighting the loss by the size of the tree below it? $\endgroup$
    – Emre
    May 3, 2016 at 17:25
  • $\begingroup$ No, I haven't tried that. Could you please tell me more about how to do it? $\endgroup$
    – nino
    May 4, 2016 at 7:57


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.