Title :
Properties of deletion methods in competitive learning
Author :
Macda, M. ; Miyajima, Hiromi
Author_Institution :
Kurume Nat. Coll. of Technol., Japan
Abstract :
In this paper, we describe properties of deletion methods in competitive learning. From the viewpoint of the deleting mechanisms of reference vectors, we introduce approaches termed the adaptivity and sensitivity deletions participating in the criteria of partition error and distortion error, respectively. Experimental results show the effectiveness of the present approaches in the average distortion
Keywords :
errors; unsupervised learning; vectors; adaptivity deletions; competitive learning; deletion methods; distortion error; partition error; reference vectors; sensitivity deletions; Petroleum;
Conference_Titel :
Circuits and Systems, 2001. ISCAS 2001. The 2001 IEEE International Symposium on
Conference_Location :
Sydney, NSW
Print_ISBN :
0-7803-6685-9
DOI :
10.1109/ISCAS.2001.921430
