English

Matrix Co-completion for Multi-label Classification with Missing Features and Labels

Machine Learning 2018-05-24 v1 Machine Learning

Abstract

We consider a challenging multi-label classification problem where both feature matrix \X\X and label matrix \Y\Y have missing entries. An existing method concatenated \X\X and \Y\Y as [\X;\Y][\X; \Y] and applied a matrix completion (MC) method to fill the missing entries, under the assumption that [\X;\Y][\X; \Y] is of low-rank. However, since entries of \Y\Y take binary values in the multi-label setting, it is unlikely that \Y\Y is of low-rank. Moreover, such assumption implies a linear relationship between \X\X and \Y\Y which may not hold in practice. In this paper, we consider a latent matrix Z\Z that produces the probability σ(Zij)\sigma(Z_{ij}) of generating label YijY_{ij}, where σ()\sigma(\cdot) is nonlinear. Considering label correlation, we assume [\X;Z][\X; \Z] is of low-rank, and propose an MC algorithm based on subgradient descent named co-completion (COCO) motivated by elastic net and one-bit MC. We give a theoretical bound on the recovery effect of COCO and demonstrate its practical usefulness through experiments.

Keywords

Cite

@article{arxiv.1805.09156,
  title  = {Matrix Co-completion for Multi-label Classification with Missing Features and Labels},
  author = {Miao Xu and Gang Niu and Bo Han and Ivor W. Tsang and Zhi-Hua Zhou and Masashi Sugiyama},
  journal= {arXiv preprint arXiv:1805.09156},
  year   = {2018}
}
R2 v1 2026-06-23T02:05:44.100Z