Title :
DGTSOM: An Improved Dynamical Growing Tree Based on Self-Organizing Map
Author :
Zhang, Qian ; Qi, Deyu
Author_Institution :
Res. Inst. of Comput. Syst., South China Univ. of Technol., Guangzhou, China
Abstract :
The advantage and disadvantage of some kinds of the improved self-organizing map algorithm are discussed in the paper, and an Improved Dynamical Growing Tree based on Self-organizing Map (DGTSOM) is introduced. In the proposed algorithm, the network size and shape is formed by growing nodes on demand in the right position and pruning underused nodes during the unsupervised training process, so the network structure is flexible and dynamical, not needed to be predetermined. The DGTSOM algorithm is presented in detail, and the performance advantages are discussed and compared with other algorithms.
Keywords :
pattern classification; self-organising feature maps; tree data structures; unsupervised learning; DGTSOM algorithm; improved dynamical growing tree; improved self-organizing map algorithm; pruning underused nodes; unsupervised training process; Accuracy; Classification algorithms; Clustering algorithms; Convergence; Equations; Heuristic algorithms; Training;
Conference_Titel :
Pattern Recognition (CCPR), 2010 Chinese Conference on
Conference_Location :
Chongqing
Print_ISBN :
978-1-4244-7209-3
Electronic_ISBN :
978-1-4244-7210-9
DOI :
10.1109/CCPR.2010.5659328