Random block coordinate descent method for multi-label support vector machine with a zero label
作者：
Highlights：
• A random block coordinate descent method for multi-label SVM with a zero label is proposed.
• Theoretical analysis shows that our novel method has a lower time complexity.
• This novel training procedure runs averagely 11 times faster than the original algorithm based on Frank–Wolfe method.
• This method produces fewer support vectors compared with Frank–Wolfe method.
• Our new algorithm is an effective and efficient candidate for multi-label classification.
摘要
•A random block coordinate descent method for multi-label SVM with a zero label is proposed.•Theoretical analysis shows that our novel method has a lower time complexity.•This novel training procedure runs averagely 11 times faster than the original algorithm based on Frank–Wolfe method.•This method produces fewer support vectors compared with Frank–Wolfe method.•Our new algorithm is an effective and efficient candidate for multi-label classification.
论文关键词：Multi-label classification,Support vector machine,Zero label,Frank–Wolfe method,Block coordinate descent method,Quadratic programming,Linear programming
论文评审过程：Available online 9 December 2013.
论文官网地址：https://doi.org/10.1016/j.eswa.2013.12.004
