In many supervised learning problems, it is assumed that data is independent and identically distributed. This?assumption does not hold true in many real cases, where a neighboring pair of examples and their labels exhibit?some kind of relationship. Sequential learning algorithms take benefit of these relationships in order to improve?generalization. In the literature, there are different approaches that try to capture and exploit this correlation?by means of different methodologies. In this thesis we focus on meta-learning strategies and, in particular, the?stacked sequential learning (SSL) framework. The main contribution of this thesis is to generalize the SSL highlighting the key role of how to model the neighborhood interactions. We propose an effective and efficient way of capturing and exploiting sequential correlations that take into account long-range interactions. We tested our method on several tasks: text line classification, image pixel classification, multi-class classification problems and human pose segmentation. Results on these tasks clearly show that our approach outperforms the standard stacked sequential learning as?well as off-the-shelf graphical models such conditional random fields.
展开▼
