Self-Paced Multi-Label Learning with Diversity

  • 2019-10-08 15:59:57
  • Seyed Amjad Seyedi, S. Siamak Ghodsi, Fardin Akhlaghian, Mahdi Jalili, Parham Moradi
  • 0

Abstract

The major challenge of learning from multi-label data has arisen from theoverwhelming size of label space which makes this problem NP-hard. This problemcan be alleviated by gradually involving easy to hard tags into the learningprocess. Besides, the utilization of a diversity maintenance approach avoidsoverfitting on a subset of easy labels. In this paper, we propose a self-pacedmulti-label learning with diversity (SPMLD) which aims to cover diverse labelswith respect to its learning pace. In addition, the proposed framework isapplied to an efficient correlation-based multi-label method. The non-convexobjective function is optimized by an extension of the block coordinate descentalgorithm. Empirical evaluations on real-world datasets with differentdimensions of features and labels imply the effectiveness of the proposedpredictive model.

 

Quick Read (beta)

This feature is not avaialbe for this paper.