一种新高斯过程分类算法

由于需要利用高斯函数逼近潜变量函数的后验概率, 传统高斯过程分类算法通常都存在计算复杂度高的问题. 对此, 提出一种新高斯过程分类算法. 该算法的基本思想为: 首先, 利用Parzen 窗方法估计出每个训练样本的后验概率; 然后, 通过所得到的后验概率将原始分类问题变换为回归问题; 进而分析地得到潜变量函数后验概率的显式表达式, 以避免逼近后验概率所面临的高计算复杂度问题. 仿真实验结果表明, 所提出的算法在分类精度上优于已有的高斯过程分类算法.

A new Gaussian process classification algorithm

Because the posterior probability of the latent function needs to be approximated by a tractable Gaussian function, the traditional Gaussian process classification algorithms usually suffer from high computational cost. Therefore, a new Gaussian process classification algorithm is proposed. The basic idea is to use Parzen-window method to estimate the posterior probability of training data, and then transform the classification problem to a regression problem based on the obtained posterior probability. As a result, the explicit expression of the posterior probability of the latent function can be derived analytically and the high computational cost caused by approximating the posterior probability with Gaussian distribution is also avoided. The experimental results show that the proposed algorithm can achieve superior classification accuracy to the existing Gaussian process classification algorithms.