中文说明：它的训练是在贝叶斯框架下进行的，在先验参数的结构下基于主动相关决策理论（automatic relevance determination，简称ARD）来移除不相关的点，从而获得稀疏化的模型。在样本数据的迭代学习过程中，大部分参数的后验分布趋于零，与预测值无关，那些非零参数对应的点被称作相关向量（Relevance Vectors），体现了数据中最核心的特征。同支持向量机相比，相关向量机最大的优点就是极大地减少了核函数的计算量，并且也克服了所选核函数必须满足Mercer条件的缺点。
English Description：
Its training is carried out under the framework of Bayesian. Under the structure of prior parameters, it is based on automatic correlation determination (ARD) to remove irrelevant points, so as to obtain a sparse model. In the iterative learning process of sample data, the posterior distribution of most parameters tends to zero, which has nothing to do with the predicted value. The points corresponding to non-zero parameters are called correlation vectors, which reflect the core characteristics of the data. Compared with support vector machine (SVM), relevance vector machine (RVM) has the advantage of greatly reducing the computation of kernel function and overcoming the shortcoming that the selected kernel function must satisfy Mercer condition.