Search-based Structured Prediction - Computer Science > LearningReport as inadecuate

Search-based Structured Prediction - Computer Science > Learning - Download this document for free, or read online. Document in PDF available to download.

Abstract: We present Searn, an algorithm for integrating search and learning to solvecomplex structured prediction problems such as those that occur in naturallanguage, speech, computational biology, and vision. Searn is a meta-algorithmthat transforms these complex problems into simple classification problems towhich any binary classifier may be applied. Unlike current algorithms forstructured learning that require decomposition of both the loss function andthe feature functions over the predicted structure, Searn is able to learnprediction functions for any loss function and any class of features. Moreover,Searn comes with a strong, natural theoretical guarantee: good performance onthe derived classification problems implies good performance on the structuredprediction problem.

Author: Hal Daumé III, John Langford, Daniel Marcu


Related documents