Limited-Supervised Multi-Label Learning with Dependency Noise

• DOI: 10.1609/aaai.v38i14.29494
• Publication Date: 2024-03-25T11:31:12Z
• Authors: Yejiang Wang; Yuhai Zhao; Zhengkui Wang; Wen Shan; Xingwei Wang

Limited-supervised multi-label learning (LML) leverages weak or noisy supervision for classification model training over data with label noise, which contain missing labels and/or redundant labels. Existing studies usually solve LML problems by assuming that noise is independent of the input features and class labels, while ignoring fact may depend on (instance-dependent) classes (label-dependent) in many real-world applications. In this paper, we propose limited-supervised Multi-label Learning Dependency Noise (MLDN) to simultaneously identify instance-dependent label-dependent factorizing matrix as outputs a mapping from feature representations. Meanwhile, regularize problem manifold constraint preserve local relationships uncover structure. Theoretically, bound recover error resulting problem. We using first-order scheme based proximal operator, convergence rate it at least sub-linear. Extensive experiments conducted various datasets demonstrate superiority our proposed method.