.png){kind=icon}
返回
A safe acceleration method for multi-task twin support vector machine
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-021-01481-8.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Multi-task twin support vector machine (DMTSVM) performs better than twin support vector machine (TSVM) since it takes full advantage of the shared information among all tasks. However, it remains challenging to apply DMTSVM to large-scale problems directly. Inspired by the sparsity of DMTSVM, two safe sample screening rules SSRC-DMTSVM and SSRR-DMTSVM for DMTSVM are proposed in this paper. The two rules can identify and delete most inactive instances before solving the optimization problem. Therefore, the computational cost can be reduced greatly. More importantly, they are safe in the sense that the proposed methods could achieve the exactly same solution as solving the original problem. Besides, our screening rules are independent of the solver, thus they can be combined with other fast algorithms. To further accelerate the computational speed, we employ a dual coordinate descent method (DCDM) for DMTSVM. Numerical experiments on eight synthetic data sets, nine multi-task benchmark data sets, a real Chinese grape wine data set and two popular image data sets demonstrate that the computational cost can be dramatically reduced without sacrificing the accuracy by our proposed methods.
Keyword:
Twin support vector machine
Safe screening rule
Multi-task learning
DCDM
期刊
IF:
2.7
论文数:
3.2K
被引数:
5.6K
机构
引用论文
Investigation of the High-Temperature and Rheological Properties for Asphalt Sealant Modified by SBS and Rubber Crumb.png)
Polymers
OAAI
PolymersIF0
OAAI分享收藏
查看更多内容