Partial Multi-Label Learning by Low-Rank and Sparse Decomposition

Abstract

Multi-Label Learning (MLL) aims to learn from the training data where each example is represented by a single instance while associated with a set of candidate labels. Most existing MLL methods are typically designed to handle the problem of missing labels. However, in many real-world scenarios, the labeling information for multi-label data is always redundant , which can not be solved by classical MLL methods, thus a novel Partial Multi-label Learning (PML) framework is proposed to cope with such problem, i.e. removing the the noisy labels from the multi-label sets. In this paper, in order to further improve the denoising capability of PML framework, we utilize the low-rank and sparse decomposition scheme and propose a novel Partial Multi-label Learning by Low-Rank and Sparse decomposition (PML-LRS) approach. Specifically, we first reformulate the observed label set into a label matrix, and then decompose it into a groundtruth label matrix and an irrelevant label matrix, where the former is constrained to be low rank and the latter is assumed to be sparse. Next, we utilize the feature mapping matrix to explore the label correlations and meanwhile constrain the feature mapping matrix to be low rank to prevent the proposed method from being overfitting. Finally, we obtain the ground-truth labels via minimizing the label loss, where the Augmented Lagrange Multiplier (ALM) algorithm is incorporated to solve the optimization problem. Enormous experimental results demonstrate that PML-LRS can achieve superior or competitive performance against other state-of-the-art methods.

Partial Multi-Label Learning with Noisy Label Identification

“…For example, (Xie and Huang 2018) propose two effective methods PML-lc and PML-fp by introducing a confidence value for each candidate label. The decomposition scheme is utilized to tackle PML data in (Lijuan Sun and Jin 2019). PARTICLE (Fang and Zhang 2019) identifies the credible labels with high labeling confidences by employing an iterative label propagation procedure.…”

“…In multi-label learning, a common assumption is that there existing the label correlations among different labels (Zhu, Kwok, and Zhou 2017;Lijuan Sun and Jin 2019) and the feature mapping matrix U is thus linearly dependent. The low-rank assumption is thus naturally used to capture this intrinsic property of the classifier.…”

“…Among them, the most commonly used strategy to learn from PML examples is disambiguation. It tires to recover ground-truth labeling information from candidate labels, by either introducing labeling confidences (Xie and Huang 2018;Fang and Zhang 2019) or employing low-rank and sparse decomposition scheme (Lijuan Sun and Jin 2019). Despite the advances these methods have achieved, a potential limitation is that they neglect the cause of noisy labels in the candidate set, which may be an important information for recovering the ground-truth labels.…”

A Self-Ensemble Approach for Partial Multi-Label Learning

“…However, the confidence values would be error-prone especially when noisy label dominates since it ignores the irrelevance of the non-candidate labels. Low-rank assumption is employed to disambiguate noisy labels by using sparse matrix decomposition [4]. Another recent work [5] proposes to deal with PML problem by using a two-stage strategy.…”

“…A reasonable result is achieved, while the estimation of label confidence scores is error-prone especially with the high proportion of false positive labels since it ignores the irrelevance of the noncandidate labels. Some researchers employ sparse decomposition scheme based on a low-rank assumption to identify the noise labels for disambiguation [4]. Another recent work of PML in [5] tries to estimate the label confidence scores of candidate labels by employing iterative label propagation; and then, a credible label elicitation techniques is employed to identify the credible labels which can be used to induce a predictive model for making final prediction on unseen instances.…”