TY - GEN
T1 - Rival penalization controlled competitive learning for data clustering with unknown cluster number
AU - Cheung, Yu Ming
N1 - Funding Information:
The work described in this paper was supported by the Faculty Research Grant of Hong Kong Baptist University with project number: FRG/01-02/II-24.
PY - 2002
Y1 - 2002
N2 - Conventional clustering algorithms such as k-means (Forgy 1965, MacQueen 1967) need to know the exact cluster number k∗ before performing data clustering. Otherwise, they will lead to a poor clustering performance. Unfortunately, it is often hard to determine k∗ in advance in many practical problems. Under the circumstances, Xu et al. in 1993 proposed an approach named Rival Penalized Competitive Learning (RPCL) algorithm that can perform appropriate clustering without knowing the cluster number by automatically driving extra seed points far away from the input data set. Although RPCL has made great success in many applications, its performance is however very sensitive to the selection of the de-learning rate. To our best knowledge, there is still an open problem to guide this rate selection. We further investigate RPCL by presenting a mechanism to dynamically control the rival-penalizing forces. Consequently, we give out a rival penalized controlled competitive learning (RPCCL) approach, which circumvents the selecting problem of the de-learning rate by always fixing it at the same value as the learning rate. In contrast, the RPCL cannot do that in the same way. The experiments have shown the outstanding performance of this algorithm in comparison with the RPCL.
AB - Conventional clustering algorithms such as k-means (Forgy 1965, MacQueen 1967) need to know the exact cluster number k∗ before performing data clustering. Otherwise, they will lead to a poor clustering performance. Unfortunately, it is often hard to determine k∗ in advance in many practical problems. Under the circumstances, Xu et al. in 1993 proposed an approach named Rival Penalized Competitive Learning (RPCL) algorithm that can perform appropriate clustering without knowing the cluster number by automatically driving extra seed points far away from the input data set. Although RPCL has made great success in many applications, its performance is however very sensitive to the selection of the de-learning rate. To our best knowledge, there is still an open problem to guide this rate selection. We further investigate RPCL by presenting a mechanism to dynamically control the rival-penalizing forces. Consequently, we give out a rival penalized controlled competitive learning (RPCCL) approach, which circumvents the selecting problem of the de-learning rate by always fixing it at the same value as the learning rate. In contrast, the RPCL cannot do that in the same way. The experiments have shown the outstanding performance of this algorithm in comparison with the RPCL.
UR - http://www.scopus.com/inward/record.url?scp=84965028127&partnerID=8YFLogxK
U2 - 10.1109/ICONIP.2002.1202214
DO - 10.1109/ICONIP.2002.1202214
M3 - Conference proceeding
AN - SCOPUS:84965028127
T3 - ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing: Computational Intelligence for the E-Age
SP - 467
EP - 471
BT - ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing
A2 - Rajapakse, Jagath C.
A2 - Yao, Xin
A2 - Wang, Lipo
A2 - Fukushima, Kunihiko
A2 - Lee, Soo-Young
PB - IEEE
T2 - 9th International Conference on Neural Information Processing, ICONIP 2002
Y2 - 18 November 2002 through 22 November 2002
ER -