Burg matrix divergence based multi-metric learning

The basic idea of most distance metric learning methods is to find a space that can optimally classify data points belong to different categories. However, current methods only learn one Mahalanobis distance for each data set, which actually fails to perfectly classify different categories in most real world applications. To improve the classification accuracy of k-nearest-neighbour algorithm, a multi-metric learning method is proposed in this paper to completely classify different categories by sequentially learning sub-metrics. The proposed algorithm is based on minimizing the Burg matrix divergence between metrics. The experiments on five UCI data sets demonstrate the improved performance of Multi-Metric learning when comparing with the state-of-the-art methods.