基于稀疏重构消歧的偏标记分类算法

PDF下载HTML

Title:

Partial label classification algorithm based on sparse reconstruction disambiguation

作者:

殷建华, 刘振丙, 魏黄曌

桂林电子科技大学 计算机与信息安全学院, 广西 桂林 541004

Author(s):

YIN Jianhua, LIU Zhenbing, WEI Huangzhao

School of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin 541004, China

关键词:

弱监督学习; 稀疏重构; 平滑假设; 标签消歧; 偏标记学习; 候选标签; 特征空间; 多分类

Keywords:

weakly supervised learning; sparse reconstruction; smoothness assumption; label disambiguation; partial label learning; candidate label; feature space; multi-classification

分类号:

TP181

DOI:

10.11992/tis.202202024

摘要:

针对现有的大多数方法在消歧过程中缺乏对特征空间潜在有用信息的利用和对候选标签不同置信度水平的考虑的问题，本文提出了一种基于稀疏重构消歧的偏标记学习( partial label learning by sparse reconstruction disambiguation，PL-SRD)的新方法，利用特征空间的结构信息促进标签的消歧过程。本文通过对训练样本进行稀疏重构来刻画特征空间的拓扑结构并将其融入到标签消歧过程中；提出一个统一的框架将标签消歧与训练预测模型同时进行。在人工合成和真实数据集上进行的大量实验表明，本文提出的方法比多个现有的偏标记学习算法取得了更好的性能。

Abstract:

Most of the existing partial label learning (PLL) methods neither ignores making full use of potentially useful information from feature space, nor considers different labeling confidence levels of the candidate labels in the disambiguation process. In this paper, we propose a novel approach for partial label learning by sparse reconstruction disambiguation (PL-SRD), which facilitates the labeling disambiguation process by leveraging the structural information in feature space. We first characterize the topological structure of feature space by conducting sparse reconstruction among the training examples and integrate it into the label disambiguation process. Then, we present a unified framework, which performs label disambiguation and predictive model training simultaneously. Extensive experimental results on both artificial and real-world datasets demonstrate the superiority of our method to other state-of-the-art PLL methods.