摘要
为降低多示例学习中噪声示例对分类结果的影响,提出赋予包示例不同的权值,不断更新分类器,调整权值,提高分类精度。在传统的多示例学习中,训练集由若干个包组成,每个包包含若干个示例,包示例标签未知。受获取数据的环境和传输过程等不确定因素的影响,现实世界的数据极易受到噪声的干扰,在多示例学习中,正包中存在正示例,也可能包含负示例噪声,这些噪声会影响分类效果。实验结果表明,该方法具有更好的分类能力。
To decrease the influence of noise in multiple instance learning,all instances in bags with different weights were proposed,the classifier was updated,weights were adjusted to improve classification accuracy.In traditional multiple instance learning,the training set is composed of a set of bags,each bag contains many instances,and instance label is unknown.Because of many uncertain factors,such as the acquisition environment,the transmission process and so on,the data are influenced by the noise.Also,positive bag may contain negative instance noise and not all instances in it are positive,results of classification are affected by noise.Experimental results show that the proposed algorithm has better classification capability.
出处
《计算机工程与设计》
北大核心
2016年第5期1271-1274,共4页
Computer Engineering and Design
基金
国家自然科学基金项目(61472089
61202270)
关键词
多示例学习
噪声
示例加权
稀疏正包
分类
multiple instance learning
noise
instance weighted
sparse positive bags
classification
