多示例学习的包层次覆盖k近邻算法.pdfVIP

小型微型计算机系统 Journal of Chinese Computer Systems 2014 年 11 月第 11 期 Vol. 35 No. 11 2014 多示例学习的包层次覆盖 k 近邻算法 赵妹l气芮辰1气陈洁1气张燕平1 ，2 1 (安徽大学计算智能与信号处理教育部重点实验室，合肥 23侃01) 2( 安徽大学人工智能研究所，合肥 23侃01) E-mail:zhangyp2@ gmail. ωm 摘 要:多示例学习是一种新型的机器学习框架，正包中大量的噪声使多示例数据集具有很大的歧义性.为了排除多示例数据 集正包中大量的假正例，提高分类精度，结合邻域覆盖算法，提出一个新的多示例包层次覆盖 k 近邻算法.覆盖算法的学习结果 是一系列的球形邻域，在每一个球形邻域中只含有同类样本，本文利用的覆盖算法的这一特性重新组织多示例数据集的包结 构.概括的说，为了排除正包中大量的假正例，首先对原有的多示例包结构进行重新构造，使用覆盖算法生成的球形邻域做为新 的包结构，从而提高多示例样本在新的特征空间中的可分离性.然后，使用包层次的 k 近邻算法排除正包中大量的噪声并预测 测试包的类别.实验表明，多示例学习的包层次覆盖 k 近邻算法具有很好的性能. 关键词:机器学习;多示例学习;覆盖算法;k 近邻算法 中固分类号:币18 文献标识码:A 文章编号: l000-1220( 2014) 11-2511 -04 Multi-instance Learning Bag-Ievel Covering-kNN A1gorithm ZHAO ShU I ,2 , RUI Chen l ,2 ， Cl西N Jiel ,2 ,ZHANG Yan-pingl ,2 1 ( Key Lab of lntelligent Computing Signal 阶ocessing ， Ministry of Education , Anhui University , Hefei 23创始I ， China) 2 ( lnstitute of Artificial lntelligence , Ar由ui University , Hetie 23仪坦白 I ，Cbina) Abs仕act:Multi-instance learning is a new 仕缸nework in machine learning. An extensive number of noises in the positive bags is 也e inherent difficulty of mulι?llstance learning. In order to improve the classification accuracy ，也is paper puts forw缸d a novel multi-in- stance learning bag-level Covering-kNN algori也m to exclude the noises in multi-instance data set.ηle learning results of Covering al- gori也m is a set of sphere neighbors and each sphere neighbor only contains 阴阳ms belong to the same class.η山 f饵ture help us re- org缸咀ze the s位ucture of bags in multi-instance 也.ta set. Generally sp臼挝晤， in order to exclude false positive instances in the positive bags , frrst , we reconstruct 由e structure of multi-instance data set by treating the sphere neighbors obtained using Covering algorithm as 由e new struc阳re of bags. Thus , improving 由e separable of multi-instance samples in 由e new feature space. Then , the bag-level kNN algorithm is utilized to exclude the noises in positive bags and predict the labels of test bags. The experiments demon