New Method Focused on Metric Learning Accepted to IJCAI 2019

Tue May 14, 2019

Kai Liu and Lodewijk Brand have recently had their paper, “Learning Robust Distance Metric with Side Information via Ratio Minimization of Orthogonally Constrained L2,1-Norm Distances”, accepted into the 28th International Joint Conference on Artificial Intelligence (IJCAI 2019).

This work studies a distance metric learning method by utilizing side information from must-link and cannot-link datasets. Instead of traditional squared l2 norm distances, an efficient iterative algorithm is proposed to solve the general min-max objectives that use l21-norm, while yielding a strictly orthogonal projection matrix. Extensive experiments are conducted, where the proposed distance metric learning method outperforms related state-of-the art methods in a variety of experimental settings.

This paper will be presented this August in Macao, China. IJCAI is a top conference in Artificial Intelligence.