A New SOM Initialization Algorithm for Nonvectorial Data

Research areas:

• Uncategorized

Year:

2008

Type of Publication:

In Proceedings

Keywords:

Median SOM, initialization, pairwise data

Authors:

• Antonino Fiannaca
• Riccardo Rizzo
• Alfonso Urso
• Salvatore Gaglio

Volume:

5177

Book title:

Knowledge-Based Intelligent Information and Engineering Systems

Pages:

41-48

ISBN:

978-3-540-85562-0

ISSN:

0302-9743

Abstract:

Self Organizing Maps (SOMs) are widely used mapping and clustering algorithms family. It is also well known that the performances of the maps in terms of quality of result and learning speed are strongly dependent from the neuron weights initialization. This drawback is common to all the SOM algorithms, and critical for a new SOM algorithm, the Median SOM (M-SOM), developed in order to map datasets characterized by a dissimilarity matrix. In this paper an initialization technique of M-SOM is proposed and compared to the initialization techniques proposed in the original paper. The results show that the proposed initialization technique assures faster learning and better performance in terms of quantization error.