Abstract

Kernel-Based Discriminative Learning Algorithms for Labeling Sequences, Trees, and Graphs
Hisashi Kashima - IBM Tokyo Research Laboratory Yuta Tsuboi - IBM Tokyo Research Laboratory
We introduce a new perceptron-based discriminative learning algorithm forlabeling structured data such as sequences, trees, and graphs. Since it is fully kernelized and uses pointwise label prediction, large features, including arbitrary number of hidden variables, can beincorporated with polynomial time complexity. This is in contrast to existing labelers that can handle only features of asmall number of hidden variables, such as Maximum Entropy Markov Models and Conditional Random Fields. We also introduce several kernel functions for labeling sequences, trees andgraphs and efficient algorithms for them.

Kernel-Based Discriminative Learning Algorithms for Labeling Sequences, Trees, and Graphs

Hisashi Kashima - IBM Tokyo Research Laboratory
Yuta Tsuboi - IBM Tokyo Research Laboratory

We introduce a new perceptron-based discriminative learning algorithm forlabeling structured data such as sequences, trees, and graphs. Since it is fully kernelized and uses pointwise label prediction, large features, including arbitrary number of hidden variables, can beincorporated with polynomial time complexity. This is in contrast to existing labelers that can handle only features of asmall number of hidden variables, such as Maximum Entropy Markov Models and Conditional Random Fields. We also introduce several kernel functions for labeling sequences, trees andgraphs and efficient algorithms for them.