欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
为了克服传统的分类器难以在具有令人满意的分类性能、快速的学习效率的同时兼顾高可解释性之不足,提出增量式0阶模糊分类器TSK-IFC0IRLS.该分类器通过使用增量式模糊聚类算法IFCM（c＋p）对训练样本进行聚类,使用高斯隶属度函数将聚类结果映射到模糊子空间,使用迭代重加权最小二乘优化算法IRLS对模糊规则的后件参数进行学习.通过提出基于伪Huber函数的代价函数,它的鲁棒性改进版本TSK-IFC0PHub被提出来以提高分类器的抗噪能力.仿真实验表明,与FCPM-IRLS、RBF、ANFIS分类器相比,提出的2种模糊分类器均具有良好的分类性能及数据规模的可扩展性,TSK-IFC0PHub具有良好的鲁棒性.
英文摘要：
An incremental zero-order TSK fuzzy classifier called TSK-IFC0 IRLSwas proposed based on iteratively reweighted least squares optimization algorithm in order to circumvent the drawback that traditional non-fuzzy classifiers had no any interpretability and that fuzzy classifiers could not always be feasible for many datasets with satisfied classification performance.The incremental fuzzy clustering algorithm IFCM（c＋p）for large-scale datasets was used to quickly train antecedent parameters of fuzzy rules by clustering and using Gauss function to map the clustering results into fuzzy subspace.The iteratively reweighted least squares optimization algorithm was used to learn consequent parameters of fuzzy rules.The robust version called TSK-IFC0 PHubwas developed based on pseudo-Huber loss function with the purpose of improving anti-noise ability of TSK-IFC0 IRLS.The proposed fuzzy classifiers were experimentally compared with conventional fuzzy classifier FCPM-IRLS,RBF neural network and ANFIS.Results indicated the power of the proposed fuzzy classifiers on interpretability,classification performance and scalability.The strong robust capability of TSK-IFC0 PHubwas verified by the experimental results.
同期刊论文项目
同项目期刊论文
Dm-KDE: dynamical kernel density estimation by sequences of KDE estimators with fixed number of comp
An Improved TA-SVM Method Without Matrix Inversion and Its Fast Implementation for Nonstationary Dat
T2FELA: type-2 fuzzy extreme learning algorithm for fast training of interval type-2 TSK fuzzy logic
Generalized hidden-mapping ridge regression, knowledge-leveraged inductive transfer learning for neu
Minimax probability TSK fuzzy system classifier: a more transparent and highly interpretable classif
期刊信息
位置：