Boosting Algorithm with Sequence-Loss Cost Function for Structured Prediction

摘要

The problem of sequence prediction i.e. annotating sequences appears in many problems across a variety of scientific disciplines, especially in computational biology, natural language processing, speech recognition, etc. The paper investigates a boosting approach to structured prediction, AdaBoost~(STRUCT), based on proposed sequence-loss balancing function, combining advantages of boosting scheme with the efficiency of dynamic programming method. In the paper the method's formalism for modeling and predicting label sequences is introduced as well as examined, presenting its validity and competitiveness.