摘要
增量学习是一种在巩固原有学习成果的基础上快速有效地获取新知识的学习模式 .本文在简述增量学习的相关研究以及排序学习前向掩蔽模型 (SLAM)的特点后 ,提出了一种基于SLAM的快速增量学习算法 .该算法在原神经网络模型分类能力的基础上 ,实现对新增样本的快速增量学习 ,从而在较短的时间内提高该网络模型的分类推广能力 .最后 ,与SLAM算法和Levenberg Marquardt后向传播 (LMBP)
Incremental learning mode is meaningful to efficiently acquire additional knowledge on the basis of original knowledge structure. In this paper, the research background of incremental learning and the characteristics of sequential learning ahead masking model (SLAM) are firstly described. Then the fast incremental learning algorithm based on sequential learning ahead masking model is presented. Finally, compared with the SLAM algorithm and Levenberg-Marquardt Back Propagation algorithm, the proposed algorithm is testified to have the better computation efficiency and generalization ability.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2004年第12期2051-2055,共5页
Acta Electronica Sinica
基金
国家自然科学基金 (No .30 2 30 350 )
关键词
排序学习前向掩蔽模型
增量学习
神经网络
Convergence of numerical methods
Database systems
Knowledge engineering
Knowledge representation
Learning algorithms
Mathematical models
