四种TSVR型学习算法的性能比较

Performance Comparison for Four TSVR-type Learning Algorithms

我们知道,基于SVR的学习算法的计算复杂性和稀疏性对分析和处理大数据来说是非常重要的两个因素,尤其是对高维数据.为此,学者们做了大量的研究工作并提出了许多改进的SVR型算法.它们当中,有些算法的出发点基本相同,只是求解方法上略有不同;有些算法有明显不同的出发点,其所构建的最优化模型也不相同,但求解方法上大同小异.本文选择四个较具代表性的TSVR型学习算法,分析和比较它们的性能,以期更加深入的理解这些算法,且在应用中更具有选择性.

It is w ell know n that the computational complexity and sparsity of learning algorithms based on support vector regression machines （SVRs） are two main factors for analyzing and treating big data ,especially for high dimensional data .According to the two factors ,scholars did a lot of research work and proposed many improved SVR‐type learning algorithms .Among these improved algorithms , some have the basically same starting point ,just solving methods are slightly different ;some have dis‐tinctly different starting point and then result in different optimization problems ,but the solving meth‐ods are similar .For deep understanding these improved algorithms and being more selective in the appli‐cations ,this paper is devoted to analyze and compare the performance for four more representative TS‐VR‐type algorithms .