Home // COGNITIVE 2015, The Seventh International Conference on Advanced Cognitive Technologies and Applications // View article

A Dynamic GSOM-based Concept Tree for Capturing Incremental Patterns

Authors:
Pin Huang
Susan Bedingfield
Damminda Alahakoon

Keywords: growing self organizing map; clustering; concept formation; incremental learning

Abstract:
The Growing Self Organizing Map (GSOM) has been proposed to address the need of predefining network size and shape in traditional Self Organizing Maps (SOM). In the work described in this paper, the GSOM is used as a foundation for generating hierarchies of concepts in a tree structure which also has the ability to adapt and accumulate new information in an incremental learning architecture. GSOMs are used to capture inputs in time windows and the GSOM nodes are used as the base for developing the bottom level concepts in the tree. A new algorithm is then used to integrate similar information into concepts based on attribute similarities. As new data is introduced, new GSOMs are created and used to capture topological patterns which are integrated into the existing concept tree incrementally. The updated concept tree can capture multiple dimensional inputs with multi-parent nodes. It is proposed that this is an ideal building block to implement the columnar architecture in the human neo-cortex as an artificial model which could then be used as a cognitive architecture for data mining and analysis. The adaptive concept tree model is demonstrated with several benchmark data sets.

Pages: 72 to 80

Copyright: Copyright (c) IARIA, 2015

Publication date: March 22, 2015

Published in: conference

ISSN: 2308-4197

ISBN: 978-1-61208-390-2

Location: Nice, France

Dates: from March 22, 2015 to March 27, 2015