Yandongl writeup of Cohen and Carvalho

This is a review of Cohen_2005_stacked_sequential_learning by user:Yandongl.

This paper talks about sequence partitioning learning. THe idea of sacked sequential learning is built upon any existing learner to tackle sequential learning problems. MEMM low performance in this task is due to the over weight of on history data in training process while training data and test data might not be exactly consistent.

Experiments showed that this stacked sequential learning idea can improve current learner's performance. When applied to maximum entropy it outperformed MEMM, when applied to SVM it beats SVM too.