Finished Projects
Structure-Perceptron Learning of Hierarchical Object Model
(Published in CVPR 2008 [pdf])This work is part of Long Zhu’s PhD thesis research. We use a recursive multi-level model to combine the shape of object boundary and appearance of image patches. A structure-perceptron learning algorithm is proposed to learn the parameters of the model. The method performed well on the task of shape matching, foreground/background segmentation on horse and human face images.
Generation model for topic-relevant opinion document re-ranking
(Published in SIGIR 2008 [pdf])We use a generation model to explain the process of expressing opinions in text, and end up with a simple document re-ranking formula that combines topic-related information and author’s subjective polarity. Considerable improvements were obtained on TREC 2006 & 2007 blog search task.
