Abstract: Multi-instance multi-label learning is a new machine learning framework. In MIML framework, an example is described by multiple instances and associated with multiple class labels. Algorithm MIMLSVM+ decomposes the MIML problem into multiple independent binary classification problems. However, the degeneration process may lose information, which will influence the classification performance. Algorithm E-MIMLSVM+ by using multitask learning techniques is utilized to incorporate label correlations to improve the algorithm MIMLSVM+. In order to make full use of the unlabeled samples to improve the classification accuracy, this paper improves the E-MIMLSVM+ algorithm by using the semi-supervised support vector machine TSVM. In this paper, the algorithm is compared with other MIML algorithms. The experimental results show that the algorithm has achieved good classification results.

Key words: machine learning, Multi-Instance Multi-Label（MIML）, Support Vector Machine（SVM）, semi-supervised learning

摘要： 多示例多标记是一种新的机器学习框架，在该框架下一个对象用多个示例来表示，同时与多个类别标记相关联。MIMLSVM+算法将多示例多标记问题转化为一系列独立的二类分类问题，但是在退化过程中标记之间的联系信息会丢失，而E-MIMLSVM+算法则通过引入多任务学习技术对MIMLSVM+算法进行了改进。为了充分利用未标记样本来提高分类准确率，使用半监督支持向量机TSVM对E-MIMLSVM+算法进行了改进。通过实验将该算法与其他多示例多标记算法进行了比较，实验结果显示，改进算法取得了良好的分类效果。

关键词: 机器学习, 多示例多标记, 支持向量机（SVM）, 半监督学习