3网络外文.docVIP

Multi-Instance Multi-Label Learning with Weak Label? Shu-Jun Yang, Yuan Jiang, Zhi-Hua Zhou National Key Laboratory for Novel Software Technology Nanjing U niversity, Nanjing 210023, China {yangsj, jiangy, zhouzh}@lamda.nju.edu.cn 多实例多标记学习与弱标签* 蜀君阳，元江，周志华 计算机软件新技术国家重点实验室 南京理工大学，南京210023 ，中国 { yangsj ， jiangy ， zhouzh } @ lamda.nju.edu.cn 姓名： 班级：c1152 学号：19 Abstract Multi-Instance Multi-Label learning (MIML) dealswith data objects that are represented by a bag ofinstances and associated with a set of class labelssimultaneously. Previous studies typically assumethat for every training example, all positive labelsare tagged whereas the untagged labels are all neg-ative. In many real applications such as image an-notation, however, the learning problem often suffers from weak label; that is, users usually tag onlya part of positive labels, and the untagged labels arenot necessarily negative. In this paper, we propose the MIMLwel approach which works by assumingthat highly relevant labels share some common instances, and the underlying class means of bags foreach label are with a large margin. Experiments val-idate the effectiveness of MIMLwel in handling theweak label problem. 摘要 多实例多标签学习 (MIML) 处理的数据对象，它们表示的实例一袋和同时与一组类标签相关联。以前的研究通常假设为每个培训的示例中，所有正面标签标签而未加标签的标签则均为负值。在许多实际应用中例如图像表示法中，但是，学习问题经常遭受弱标签 ；就是用户通常标记仅一个部分的正面标签和未加标签的标签并不一定是负面。在此文件中，我们建议的 MIMLwel 方法，工程由假设高度相关标签分享一些常见的情况下，和基础类的袋为每个标签均与大距。实验验证 MIMLwel 的有效性在处理弱标签问题。 1 Introduction Conventional supervised learning often assumes that an ob-ject is represented by a single-instance and associated withone class label. In a recent learning framework, MultInstance Multi-Label learning (MIML) [Zhou and Zhang, 2006; Zhou et al., 2012], an object is allowed to be repre-sented by a bag of instances and associated with multiple class labels simultaneously. This framework has been fou