Daume et al, ML 2009

From Cohen Courses
Jump to navigationJump to search

Citation

Duame, H., Langford, J., and Marcu, D. 2009. Search-based structured prediction. Machine Learning. 75. 3. p297-325

Online version

Search-based structured prediction

Summary

This journal paper introduces SEARN, a meta-algorithm that combines searching and learning to make structured predictions. Note that this is the journal version of the 2006 paper that introduced this method.

The algorithm is summarized in the following figure, from page 6 of the paper:

Searn-algorithm.png

The key points from the paper are:

  • SEARN is an algorithm that solves complex structured predictions with minimal assumptions on the structure of the output and loss function
  • Their experiments show that SEARN is competitive with existing standard structured prediction algorithms on sequence labeling tasks.
  • Authors described the use of SEARN on text summarization, which yielded state-of-the-art performance.

Related papers

  • SEARN in Practice: This unpublished manuscript showcases three example problems where SEARN can be used - Daume_et_al,_2006.