论文信息 - An Improved Competitive and Cooperative Learning Approach for Data Clustering

An Improved Competitive and Cooperative Learning Approach for Data Clustering

The recently proposed Competitive and Cooperative Learning algorithm(CCL) (Cheung 2004) has provided a promising way to perform the data clustering without know- ing the number of clusters. Nevertheless, its performance is somewhat sensitive to the initialization of seed points. Also, its cooperative mechanism is applicable to the homogenous clusters only. In this paper, we will therefore suggest using the FSCL algorithm to initialize the seed points such that each cluster of data will at least have a seed point. Fur- thermore, we update the cooperation radius of seed points in CCL, whereby the improved CCL (ICCL for short) can be applicable to the heterogeneous clusters as well. Exper- iments show the efficacy of the proposed algorithm.

Yiu-ming Cheung | Wenjiang Pei | Shao-ping Wang

[1] Stanley C. Ahalt,et al. Competitive learning algorithms for vector quantization , 1990, Neural Networks.

[2] J. MacQueen. Some methods for classification and analysis of multivariate observations , 1967 .

[3] Anthony Stefanidis,et al. Self-organised clustering for road extraction in classified imagery , 2001 .

[4] Ya-Qin Zhang,et al. Vector-based signal processing and quantization for image and video compression , 1995, Proc. IEEE.

[5] John Moody,et al. Fast Learning in Networks of Locally-Tuned Processing Units , 1989, Neural Computation.

[6] Erkki Oja,et al. Rival penalized competitive learning for clustering analysis, RBF net, and curve detection , 1993, IEEE Trans. Neural Networks.

[7] Yiu-ming Cheung. A competitive and cooperative learning approach to robust data clustering , 2004, Neural Networks and Computational Intelligence.