摘要
领域适应学习旨在利用源领域中带标签的样本来解决目标领域的学习问题,其关键在于如何最大化地减小领域间的分布差异,有效解决领域间数据分布的变化。对当前领域适应学习算法进行了归纳和分类,总结了每类算法的特点,分析了5个相关典型算法并比较了其性能。最后指出了领域适应学习值得进一步探索的方向。
Domain adaptation learning aims to solve the learning problem of target domain by using the labeled samples of source domain. The key challenge is how to minimize the distribution distance among different domains at most and solve the change of data distribution effectively. Domain adaptation learning algorithms were summed up and classified. The characteristics of each type learning algorithm were summarized. Five typical algorithms were carefully analyzed and their performances were compared. What directions are worthy of further exploration was indicated.
出处
《计算机科学》
CSCD
北大核心
2015年第10期7-12,34,共7页
Computer Science
基金
国家八六三高技术研究与发展计划基金项目(2012AA01A510)
国家博士后基金项目(2013M542425)资助
关键词
领域适应学习
最大均值差
实例加权
特征映射
Domain adaptation learning, Maximum mean discrepancy, Instance weighting, Feature mapping