Field Support Vector Machines

Conventional classifiers often regard input samples as identically and independently distributed (i.i.d.). This is however not true in many real applications, especially when patterns occur as groups (where each group shares a homogeneous style). Such tasks are also called field classification. By breaking the i.i.d. assumption, one novel framework called Field Support Vector Machine (F-SVM) is proposed in this paper. The distinction lies that it is capable of training and predicting a group of patterns (i.e., a field pattern) simultaneously. Specifically, the proposed F-SVM classifier is investigated by learning simultaneously both the classifier and the Style Normalization Transformation for each group of data (called field). Such joint learning proves even feasible in the high-dimensional kernel space. An efficient optimization algorithm is further developed with the convergence guaranteed. More importantly, by appropriately exploring the style consistency in each field, the F-SVM is able to significantly improve the classification accuracy. A series of experiments are conducted to verify the effectiveness of the F-SVM model. Empirical results show that the proposed F-SVM achieves in three different benchmark data sets the best performance so far, significantly better than those state-of-the-art classifiers.

Duke Scholars

Author Kaizhu Huang DKU Faculty