[论文解读] Multi-label Learning with Missing Labels using Mixed Dependency Graphs

Baoyuan Wu, Fan Jia+3 more|arXiv (Cornell University)|Mar 31, 2018

Text and Document Classification Technologies参考文献 44被引用 5

一句话总结

该论文提出了 MLMG-CO 和 MLMG-SL 两种凸优化框架，用于在缺失标签的情况下进行多标签学习，利用混合依赖图整合实例级相似性、语义标签层次结构、类别共现关系以及低秩/稀疏标签矩阵分解。该方法在基准数据集上的图像标注和基于标签的图像检索任务中实现了最先进性能，通过结构化标签依赖建模展现出对标签稀疏性的强鲁棒性。

ABSTRACT

This work focuses on the problem of multi-label learning with missing labels (MLML), which aims to label each test instance with multiple class labels given training instances that have an incomplete/partial set of these labels. The key point to handle missing labels is propagating the label information from provided labels to missing labels, through a dependency graph that each label of each instance is treated as a node. We build this graph by utilizing different types of label dependencies. Specifically, the instance-level similarity is served as undirected edges to connect the label nodes across different instances and the semantic label hierarchy is used as directed edges to connect different classes. This base graph is referred to as the mixed dependency graph, as it includes both undirected and directed edges. Furthermore, we present another two types of label dependencies to connect the label nodes across different classes. One is the class co-occurrence, which is also encoded as undirected edges. Combining with the base graph, we obtain a new mixed graph, called MG-CO (mixed graph with co-occurrence). The other is the sparse and low rank decomposition of the whole label matrix, to embed high-order dependencies over all labels. Combining with the base graph, the new mixed graph is called as MG-SL (mixed graph with sparse and low rank decomposition). Based on MG-CO and MG-SL, we propose two convex transductive formulations of the MLML problem, denoted as MLMG-CO and MLMG-SL, respectively. Two important applications, including image annotation and tag based image retrieval, can be jointly handled using our proposed methods. Experiments on benchmark datasets show that our methods give significant improvements in performance and robustness to missing labels over the state-of-the-art methods.

研究动机与目标

• 解决训练样本存在不完整或缺失标签时多标签学习的挑战。
• 在统一图框架中建模多样化的标签依赖关系——包括实例相似性、语义层次结构、类别共现关系以及高阶标签结构。
• 开发凸优化公式，联合支持图像标注与基于标签的图像检索。
• 通过利用结构化标签依赖关系，提升在低标签供应场景下的鲁棒性与性能。
• 公开发布手动整理的语义层次结构以及语义增强的真值标签矩阵，供研究使用。

提出的方法

• 通过结合实例级相似性与类别共现关系的无向边，以及语义标签层次结构的有向边，构建混合依赖图（MG-CO）。
• 通过用完整标签矩阵的稀疏与低秩分解替代类别共现关系，提出 MG-SL，以建模高阶标签依赖关系。
• 提出 MLMG-CO 作为凸优化问题，采用二次平滑项建模实例相似性与类别共现关系，并通过线性约束建模语义层次结构。
• 通过引入两个辅助矩阵并施加与原始标签矩阵的等价约束，将稀疏与低秩矩阵分解整合到优化中，构建 MLMG-SL。
• 采用 ADMM 算法高效求解两种凸优化公式，确保可扩展性与收敛性。
• 采用向量化重表述优化问题，以支持凸性证明与算法稳定性。

实验结果

研究问题

• RQ1 整合多种类型的标签依赖关系——包括实例相似性、语义层次结构、类别共现关系以及低秩结构——是否能提升在缺失标签情况下的多标签学习性能？
• RQ2 标签矩阵的稀疏与低秩分解如何增强对高阶标签关系的建模能力？
• RQ3 统一框架能否在标签稀疏条件下，联合优化图像标注与基于标签的图像检索？
• RQ4 当标签部分缺失时，所提方法是否在准确率与鲁棒性方面优于最先进方法？
• RQ5 与仅使用共现关系或相似性相比，手动构建的语义层次结构在多大程度上提升性能？

主要发现

• 所提出的 MLMG-CO 与 MLMG-SL 方法在四个基准数据集上，针对缺失标签的多标签学习任务，显著优于最先进方法。
• 语义标签层次结构与类别共现关系的整合，在高标签稀疏性条件下显著提升了预测的鲁棒性。
• MLMG-SL 中的稀疏与低秩分解能有效捕捉高阶标签依赖关系，提升在不完整数据上的泛化能力。
• 两种方法均能联合支持图像标注与基于标签的图像检索，实现共享标签表示的双任务学习。
• 该方法在标签缺失情况下表现出强鲁棒性，即使仅提供少量标签，仍能保持高性能。
• 公开发布的手动构建语义层次结构与语义增强的真值标签矩阵，预计将显著促进未来多标签学习研究。