A Fast Implementation of Semi-Markov Conditional Random Fields

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Recently, Conditional Random Fields (CRF) model has been used and proved to be a good model for sequential modeling. It, however, lacks the capability of duration modeling. Therefore, some researchers introduced semi Markov Conditional Random Fields (semi-CRF) to take into account the duration distribution and showed some improvements. Nevertheless, the training algorithms for semi-CRF require quite a high complexity making semi-CRF impractical in some large-scale problems. Therefore, in this work we propose a fast implementation of the training algorithm in order to reduce the complexity required by semi-CRF. Our theoretical analysis as well as experiments’ result show a noticeable improvement in computation time, which is about ten times less than that of the original algorithm.