摘要
支持向量机是一种非常优秀的机器学习技术,求解大规模二次规划问题是训练SVM的关键。该文提出了一种改进方法,保持计算代价与优化步长之间的平衡,从而加速收敛,缩短训练时间。实验结果表明,在大数据集的情况下,该方法是十分有效的。
Support vector machine is an excellent machine learning technique and solving the very large quadratic programming (QP) optimizationproblem is the key of training SVM. This paper proposes an improved method based on the trade-off between the computational cost and the stepsize, which speeds the convergence and reduces the time of training SVM. The results of the experiments show that the improved method isparticularly effective for large learning task.
出处
《计算机工程》
CAS
CSCD
北大核心
2005年第12期10-12,107,共4页
Computer Engineering
关键词
支持向量机
平衡策略
序列最小优化
机器学习
Support vector machine
Trade-off strategy
Sequential minimal optimization
Machine learning
