正则化多任务学习的快速算法
发布时间：2018-09-13 15:34
【摘要】：正则化多任务学习(regularized multi-task learning,r MTL)方法及其扩展方法在理论研究及实际应用方面已经取得了较好的成果。然而以往方法仅关注于多个任务之间的关联,而未充分考虑算法的复杂度,较高的计算代价限制了其在大数据集上的实用性。针对此不足,结合核心向量机(core vector machine,CVM)理论,提出了适用于多任务大数据集的快速正则化多任务学习(fast regularized multi-task learning,Fr MTL)方法。Fr MTL方法有着与r MTL方法相当的分类性能,而基于CVM理论的Fr MTL-CVM算法的渐近线性时间复杂度又能使其在面对大数据集时仍然能够获得较快的决策速度。该方法的有效性在实验中得到了验证。
[Abstract]:Regularized multitasking learning (regularized multi-task learning,r MTL) method and its extension method have achieved good results in theoretical research and practical application. However, the previous methods only focus on the correlation between multiple tasks without fully considering the complexity of the algorithm. The high computational cost limits the practicability of the algorithm on big data set. Based on the kernel vector machine (core vector machine,CVM) theory, a fast regularization multitasking learning (fast regularized multi-task learning,Fr MTL (fast regularized multi-task learning,Fr MTL) method for multitasking big data set is proposed. The Fr MTL method has the same classification performance as the r MTL method. However, the asymptote linear time complexity of Fr MTL-CVM algorithm based on CVM theory can make it obtain faster decision speed when facing big data set. The effectiveness of the method is verified by experiments.
【作者单位】： 江南大学数字媒体学院;无锡职业技术学院物联网学院;
【基金】：国家自然科学基金面上项目No.61170122 教育部新世纪优秀人才支持计划No.NCET-120882 江苏省高校品牌专业建设工程资助项目No.PPZY2015C240~~
【分类号】：TP18
本文编号：2241588
[Abstract]:Regularized multitasking learning (regularized multi-task learning,r MTL) method and its extension method have achieved good results in theoretical research and practical application. However, the previous methods only focus on the correlation between multiple tasks without fully considering the complexity of the algorithm. The high computational cost limits the practicability of the algorithm on big data set. Based on the kernel vector machine (core vector machine,CVM) theory, a fast regularization multitasking learning (fast regularized multi-task learning,Fr MTL (fast regularized multi-task learning,Fr MTL) method for multitasking big data set is proposed. The Fr MTL method has the same classification performance as the r MTL method. However, the asymptote linear time complexity of Fr MTL-CVM algorithm based on CVM theory can make it obtain faster decision speed when facing big data set. The effectiveness of the method is verified by experiments.
【作者单位】： 江南大学数字媒体学院;无锡职业技术学院物联网学院;
【基金】：国家自然科学基金面上项目No.61170122 教育部新世纪优秀人才支持计划No.NCET-120882 江苏省高校品牌专业建设工程资助项目No.PPZY2015C240~~
【分类号】：TP18
【相似文献】
相关期刊论文 前1条
1 毛文涛;王海成;刘尚旺;;可自适应分组的超球多任务学习算法[J];计算机应用;2014年07期
相关博士学位论文 前1条
1 浦剑;多任务学习算法研究[D];复旦大学;2013年
相关硕士学位论文 前2条
1 黄艾青;基于叠加模型的多任务学习及应用[D];中国科学技术大学;2015年
2 金凤;多视角的构建及其在单任务学习和多任务学习中的应用[D];华东师范大学;2010年
，本文编号：2241588
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2241588.html
