KL-Divergence-Based and Manhattan Distance-Based Semisupervised Entropy-Regularized Fuzzy c-Means
Yuchi Kanzawa*, Yasunori Endo**, and Sadaaki Miyamoto**
*Shibaura Institute of Technology, 3-7-5 Toyosu, Koto, Tokyo 135-8548, Japan
**University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki 305-8573, Japan
In this paper, two types of semi-supervised fuzzy c-means algorithms are proposed. One feature of proposed algorithms is that they are based on an entropyregularized fuzzy c-means clustering algorithm, while conventional algorithms are based on standard fuzzy c-means. Another feature of proposed algorithms is that the membership updating equation can be obtained explicitly with any fuzzifier parameter value, while in conventional methods, the updating equation must be solved by some numerical method or by a numerically complex refinement with almost all fuzzifier parameters. The influence of supervisor-parameter and fuzzifier parameter on clustering results are discussed based on numerical experiments and compared to the conventional method, demonstrating the feasibility of proposed algorithms.
-  J. P. Bezdek, “Pattern Recognition with Fuzzy Objective Function Algorithms,” Plenum, New York, 1981.
-  S. Miyamoto and K. Umayahara, “Methods in Hard and Fuzzy Clustering,” in: Z.-Q. Liu and S. Miyamoto (Eds.), Soft computing and human-centered machines, Springer-Verlag Tokyo, 2000.
-  W. Pedrycz, “Algorithms of Fuzzy Clustering with Partial Supervision,” Pattern Recognition Letter, Vol.3, pp. 13-20, 1985.
-  M. Yamazaki, S. Miyamoto, and I.-J. Lee, “Semi-supervised Clustering with Two Types of Additional Functions,” Proc. 24th Fuzzy System Symp., 2E2-01, 2009.
-  M. Yamashiro, Y. Endo, Y. Hamasuna, and S. Miyamoto, “A Study on Semi-supervised Fuzzy c-Means,” Proc. 24th Fuzzy System Symp., 2E3-04, 2009.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 Internationa License.