摘要
数据经常分布在不同的地方,需要的机器资源也随着数据量的增长按比例增长,但数据的增长往往快于机器资源和机器学习上的改进。描述了元学习的基本过程和几种组合元分类器的度量尺度。元学习能够改进可观测性和精度,但同时过度强大的元学习技术也会导致冗余,低效甚至不精确的元分类器层次。分析这些方法的局限性并且提出了基于相异性的裁剪算法,证实了元学习和相关的裁剪方法的组合能取得相似的甚至更好的表现。
Along with the data expansion, machine resources required are increased in proportion due to the distribution of the data always in everywhere, but the amounts of data will likely grow in size faster than available improvements in machine resources and in machine learning. In this paper it describes the fundamental procedure of the recta-learning and the measurement scales of some component recta-classifiers. Observability and accuracy can be improved by meta- learning, but brute force meta-learuing techniques can result in large,redundant,inefficient and inaccurate meta-classifier hierarchies. A pruning 'algorithm on dissimilarity basis is presented after analyzing the limitation of these methods,and the combination of pruning algorithm and recta-classifiers demonstrate that they could obtain similar or even better performance.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第8期138-140,共3页
Computer Applications and Software
关键词
机器学习
元学习
基础学习者
裁剪算法
Machine learning Meta learning Base learner Pruning algorithm