Abstract

We describe a new sequential learning scheme called “stacked sequential learning”. Stacked sequential learning is a meta-learning algorithm, in which an arbitrary base learner is augmented so as make it aware of the labels of nearby examples. We evaluate the method on several “sequential partitioning problems”, which are characterized by long runs of identical labels. We demonstrate that on these problems, sequential stacking consistently improves the performance of non-sequential base learners; that sequential stacking often improves performance of learners (such as CRFs) that are designed specifically for sequential tasks; and that a sequentially stacked maximum-entropy learner generally outperforms CRFs.