Solving minimum cut based multi-label classification problem with semi-definite programming method

Abstract

Multi-label classification problem has emerged rapidly from more and more domains as the popularity and complexity of data nature. In this work, we proposed a framework that can solve multi-label classification problems that either there exist constraints among labels or not. Under this framework, the multi-label classification problem can be modeled as a minimum cut problem, where all labels and their correlations are represented by a weighted graph. If there exist constraints among the labels, a semi-definite programming (SDP) approach can be utilized. In the experimental evaluation, we conduct extensive study to compare the performance of our proposed SDP approach with other the state of art approaches. The results show that our approach has similar performance on all metrics compared to other approaches.