Generalized Label Enhancement with Sample Correlations

• DOI: 10.48550/arxiv.2004.03104
• Publication Date: 2020-01-01
• Authors: Qinghai Zheng; Jihua Zhu; Haoyu Tang; Xin‐Yuan Liu; Zhongyu Li; Huimin Lu
• Keywords: Sample (material); Benchmark (surveying); Feature (linguistics); Rank (graph theory); Feature vector

Recently, label distribution learning (LDL) has drawn much attention in machine learning, where LDL model is learned from labelel instances. Different single-label and multi-label annotations, distributions describe the instance by multiple labels with different intensities accommodate to more general scenes. Since most existing datasets merely provide logical labels, are unavailable many real-world applications. To handle this problem, we propose two novel enhancement methods, i.e., Label Enhancement Sample Correlations (LESC) generalized (gLESC). More specifically, LESC employs a low-rank representation of samples feature space, gLESC leverages tensor multi-rank minimization further investigate sample correlations both space space. Benefitting correlations, proposed methods can boost performance enhancement. Extensive experiments on 14 benchmark demonstrate effectiveness superiority our methods.