摘要
该文提出了一种基于支持向量数据描述的自适应多示例学习算法。该算法首先通过一种代表示例选取方法,在正、负包中分别选取代表示例,并将代表示例映射到特征空间,将多示例学习问题转化为特征空间中标准单示例机器学习问题,然后利用SVDD算法对特征映射后的训练样本集合进行训练得到分类器,再将代表示例更新与分类器训练交替迭代进行,最后用训练好的分类器对测试集进行预测。在多示例学习的COREL图像库进行实验,实验结果验证了算法的有效性。
This paper propose a new adaptive multi-instance learning algorithm based on SVDD.Through the representative instance selection method,the new algorithm firstly selects representative positive and negatives instances in positive and negative bags respectively,and then maps these representative instances to feature space,thus,transform the primary multi-instance learning problem into the standard single-instance learning problem in feature space,and apply the SVDD algorithm on the treated data sets to get the classifier,finally,updates representative instances and train the classifier alternately until convergence,then the trained classifier can predict the test sets.Perform experiment in the COREL image library,the results verify the effectiveness of the algorithm.
出处
《杭州电子科技大学学报（自然科学版）》
2012年第6期77-80,共4页
Journal of Hangzhou Dianzi University：Natural Sciences
基金
浙江省科技重大专项资助项目(C14032)
关键词
机器学习
代表示例选取
多示例学习
支持向量数据描述
machine learning
representative instance selection
multi-instance learning
support vector data description
