摘要
在一些模式识别应用中,具有类属信息的样本数量较少,此时监督学习算法会遇到小样本问题,导致分类器的识别精度大幅低于预期水平。基于叶分量分析,提出一种带监督信息的在线学习方法。该方法在训练过程进行监督学习,而在模式识别阶段能够在对输入样本进行分类的同时基于这些样本进行非监督在线学习,因此实现了监督学习与非监督学习的结合。在小本量情况下,在线学习可以弥补训练阶段监督学习的不足,仍能保证获得较高的识别精度。实验证明,该方法能够有效克服小样本问题。
In certain pattern recognition applications, because of the small size of the labelled training sample set, supervised learning methods will suffer from the Small Sample Size (3S) problem which results in classifier's unreasonably low recognition precision. Based on the Lobe Component Analysis (LCA) an online learning method with supervising clue is proposed. The method performs supervised learning in training,on the other hand, throughout the pattern recognition stage it has the ability of performing continually unsupervised online learning while the input samples are being classified, so supervised and unsupervised learning are combined together. Under the situation where only a small-sample-sized labelled training set is available, the online learning effort can make up for the insufficiency of the supervised learning performed in training stage,in return, an acceptable recognition accuracy is guaranteed. It is proven by experiments that the proposed method is efficient in overcoming the 3S problem.
出处
《计算机应用与软件》
CSCD
2009年第8期219-222,共4页
Computer Applications and Software
关键词
小样本
增量式学习
叶分量分析
在线学习
Small sample size (3S) Incremental learning Lobe component analysis (LCA) Online learning