Multilabel Structured Output Learning with Random Spanning Trees of Max-Margin Markov Networks

Mario Marchand (1) , Hongyu Su (2) , Emilie Morvant (3, 4) , Johu Rousu (2) , John Shawe-Taylor (5)

1 Department of Computer Science and Software Engineering [Québec]
2 Department of Information and Computer Science
3 LHC - Laboratoire Hubert Curien
4 IST Austria - Institute of Science and Technology [Klosterneuburg, Austria]
5 Centre for Computational Statistics and Machine Learning

We show that the usual score function for conditional Markov networks can be written as the expectation over the scores of their spanning trees. We also show that a small random sample of these output trees can attain a significant fraction of the margin obtained by the complete graph and we provide conditions under which we can perform tractable inference. The experimental results confirm that practical learning is scalable to realistic datasets using this approach.

Domaines: Machine Learning [stat.ML]

Mario Marchand, Hongyu Su, Emilie Morvant, Johu Rousu, John Shawe-Taylor. Multilabel Structured Output Learning with Random Spanning Trees of Max-Margin Markov Networks. Neural Information Processing Systems (NIPS), Dec 2014, Montréal, Canada. ⟨hal-01065586⟩