CAJ | 학술논문

基于现有的硬约束高斯混合模型不能处理约束违反情形，而软约束高斯混合模型又没有封闭的参数估计表达式，提出了一种基于约束正则化的生成聚类方法。该方法将约束一致正则化算子引入高斯混合模型，通过惩罚似然来处理约束违反，使满足正约束的成对样本的后验概率尽可能相似，满足负约束的成对样本的后验概率尽可能不相似；同时封闭的参数估计迭代公式降低了参数估计的计算复杂度。在一组真实数据集上的实验表明，与现有的相关方法相比，该方法能有效改善聚类性能，并对噪音约束有着更好的适应性。
기우현유적경약속고사혼합모형불능처리약속위반정형，이연약속고사혼합모형우몰유봉폐적삼수고계표체식，제출료일충기우약속정칙화적생성취류방법。해방법장약속일치정칙화산자인입고사혼합모형，통과징벌사연래처리약속위반，사만족정약속적성대양본적후험개솔진가능상사，만족부약속적성대양본적후험개솔진가능불상사；동시봉폐적삼수고계질대공식강저료삼수고계적계산복잡도。재일조진실수거집상적실험표명，여현유적상관방법상비，해방법능유효개선취류성능，병대조음약속유착경호적괄응성。
Most existing Gaussian mixture model (GMM)with hard equivalence constraints cannot solve the problem of violating pairwise constraints,and the GMM with soft equivalence constraints lacks of closed es-timation forms of model parameters.This paper presents a generative clustering analysis algorithm with con-straints regularization (GCACR),in which the constraints consistent assumption is integrating into GMM.To penalize the constraint violation,the penalized likelihood function is designed.This makes the posterior proba-bility of the pairwise data points similar or dissimilar according to the data points coming from positive con-straints or negative constraints.Meanwhile,the computational complexity of model parameter estimation is re-duced for the closed estimation forms of model parameters being provided.Experimental results on real-world datasets show that the proposed algorithm can improve the clustering performance,and can better adapt to dealing with the noise pairwise constraints compared with the state-of-the-art generative clustering algorithms.