摘要
基于高斯混合模型揭示了ML模糊推理系统构建可以等价为压缩集密度估计问题。利用此发现提出基于压缩集密度估计器RSDE的ML模糊推理系统训练算法。该算法有如下特点:①无需人为设定模糊规则数目;②是一个二次优化问题,可利用快速的二次规划算法快速求解。通过模拟和真实数据集验证,实验结果亦证实了上述优点。
Based on the Gaussian mixture Inference system (MLFIS) construction Estimation problem. Then by using this model, it is revealed that the ML (Mamdani-Larsen) Fuzzy can be equivalently taken as the Reduced Set density finding a Reduced Set density Estimator (RS.DE) based MLFIS training algorithm is presented. The proposed algorithm has the following distinctive characteristics : (1)The number of fuzzy rules is not neseccery to be set manually ; (2)It is essentially a QP propblem and can be solved directly with fast QP algorithms. The above virtues are confirmed with several experiments on synthetic and real-word datasets.
出处
《江南大学学报(自然科学版)》
CAS
2010年第1期1-6,共6页
Joural of Jiangnan University (Natural Science Edition)
基金
国家自然科学基金项目(60903100)
江苏省自然科学基金项目(BK2009067)
南京大学软件新技术国家重点实验室开放课题项目(A200602)
浙江大学CAD&CG国家重点实验室开放课题项目(A0802)
关键词
ML模糊推理系统
压缩集密度估计器
高斯混合模型
ML fuzzy inference systems, reduced set density estimator, Gaussian mixture model