支持向量机训练算法综述

训练SVM的本质是解决二次规划问题,在实际应用中,如果用于训练的样本数很大,标准的二次型优化技术就很难应用。针对这个问题,研究人员提出了各种解决方案,这些方案的核心思想是先将整个优化问题分解为多个同样性质的子问题,通过循环解决子问题来求得初始问题的解。由于这些方法都需要不断地循环迭代来解决每个子问题,所以需要的训练时间很长,这也是阻碍SVM广泛应用的一个重要原因。文章系统回顾了SVM训练的三种主流算法:块算法、分解算法和顺序最小优化算法,并且指出了未来发展方向。

A Tutorial Survey of Support Vector Machine Training Algorithms

Training SVM can be formulated into a quadratic programming problem. For large learning tasks with many training examples, off-the-shelf optimization techniques quickly become intractable in their memory and time requirements. Thus, many efficient techniques have been developed. These techniques divide the original problem into several smaller sub-problems. By solving these sub-problems iteratively, the original larger problem is solved. All proposed methods suffer from the bottleneck of long training time. This severely limited the widespread application of SVM. This paper systematically surveyed three mainstream SVM training algorithms: chunking, decomposition, and sequential minimal optimization algorithms. It concludes with an illustration of future directions.