本研究基于KISS(keep it simple and stupid)算法,利用似然比测试直接为矩阵模式定义度量,解决了现有大多数度量学习算法需要经过复杂优化过程的问题。通过在似然比测试中有目的地引入矩阵正态分布,该度量无需将矩阵模式通过向量化的方法变成向量模式,因而具有如下优点:(1)能够避免维数灾难;(2)比KISS更鲁棒;(3)无需计算大矩阵的逆和特征值分解,因此计算远快于KISS算法。最终的实验验证了该算法的优势。 Based on KISS (keep it simple and stupid) algorithm, this study directly uses the likelihood ratio test to define the metric for the matrix mode, which solves the problem that most existing metric learning algorithms need to go through complex optimization process. By intentionally introducing a matrix normal distribution in the likelihood ratio test, the metric does not need to be transformed into a vector mode by a vector quantization method, and thus has the following advantages: (1) dimensional disaster can be avoided; (2) ratio KISS is more robust; (3) Computation is much faster than the KISS algorithm without having to compute the inverse and eigenvalue decomposition of large matrices. The final experiment verifies the advantages of this algorithm.