DISTANCE METRIC LEARNING BY QUADRATIC PROGRAMMING BASED ON EQUIVALENCE CONSTRAINTS

This paper introduces a new distance metric learning algorithm which uses pair-wise equivalence (similarity and dissimilarity) constraints to improve the original distance metric in lower-dimensional input spaces. We restrict ourselves to pseudometrics that are in quadratic forms parameterized by positive semi-definite matrices. Learning a pseudo distance metric from equivalence constraints is formulated as a quadratic optimization problem, and we also integrate the large margin concept into the formulation. The proposed method works in both the input space and kernel induced feature space, and experimental results on several databases show that the learned distance metric improves the performances of the subsequent classification and clustering algorithms.