SVM增量学习中的概念迁移问题及处理方法

Concept drift on SVM incremental learning and its disposals

支持向量机由于其自身的特点使其在许多应用中表现出了特有的优势,是目前研究的热点。由于标准的SVM学习算法并不直接支持增量式学习,所以研究有效的SVM增量学习方法具有重要理论意义和实用价值。对SVM增量学习中动态目标学习的有关问题进行了深入讨论,定义了静态目标学习与动态目标学习。针对动态目标学习提出了概念迁移问题,给出了SVM增量学习概念迁移的数学表达。讨论和分析了现有的SVM增量学习方法、以及目前处理SVM增量学习中概念迁移问题的方法并得出了结论。

Support vector machine reveals its own advantages in many applications for its inherent characteristics and becomes an attractive research area these years. The standard algorithm of support vector machine cannot support incremental learning, therefore, researches on the method of effective incremental learning are of theoretical and practical important. The problem of learning on moving target in incremental learning is discussed. After giving the definition of static target learning and moving target learning, the problem of concept drift for learning on moving target is proposed, and the expression of concept drift for SVM-based incremental learning is given. The approaches of SVM-based incremental learning are discussed and analyzed, and the conclusions are given after analyzing the current approaches of disposing the concept drift on SVM-based incremental learning.