中文摘要：

针对支持向量机（SVM）适应性学习过程中产生的知识积累问题,提出了基于分类信息制导的知识积累方法.该方法为每个样本产生了用误分率表示的分类信息,误分率低的样本作为保留样本用于知识积累、参与训练和更新SVM.输入空间先被映射到高维特征空间,在该空间中依次为每个映射样本构造一个合适的分离超平面,使其对训练样本具有较低误分率,该误分率即作为相应样本包含的分类信息.与基于距离计算的传统方法相比,该方法对目标样本的定位准确性大为提高,增强了SVM的鲁棒性,提高了其适应能力.实验比较表明,该方法在计算效率和分类性能上具有非常突出的优势.

英文摘要：

Support vector machine （SVM） should accumulate its knowledge to learn new knowledge in adaptive learning prosses. Thus a knowledge accumulation method directed by classification informa tion is proposed. This method can generate the classification information represented by error separa ting rate for each sample, and those samples with lower error separating rate are kept to be as the knowledge to train and update support vector machine. The input space is firstly mapped into the high dimensional feature space, and an appropriale separating hyperplane with a lower error separating rate for the training samples is constructed for each of the mapped samples. The error separating rate is taken as the classification information of the corresponding sample. Compared with the traditional method based on distance computation, our method has higher accuracy for locating target samples, and enhances robustness of SVM and improves its adaptive abilities. Comparative numerical experi ments show that the proposed algorithm is superior in computing efficiency and classification performance.