Online Adaptive Decision Trees: Self-Organization

Recently we have shown that decision trees can be trained in the online adaptive (OADT) mode for supervised pattern classification and function approximation leading to better generalization score. The better generalization of OADT was mainly attributed to the unique property that the sum of its leaf node activations is always a constant. In this paper, we show that OADT can be used in the unsupervised mode leading to self-organization within a tree structure. We provide an architecture based on OADT, SOOADT (self-organized OADT), where we extend the basic adaptive tree with a code formation layer. The SOOADT adapts the tree parameters as well as the codes together in the online adaptive mode based on the minimization of an objective functional. We also discuss that the behavior of SOOADT is governed by a control parameter. We do not incorporate any competitive learning in SOOADT, rather we utilize the property of a constant sum of leaf node activations in obtaining the self-organizing behavior of SOOADT. Experimentally, we demonstrate the behavior of the SOOADT in identifying different groups from the data in the adaptive mode, and subsequently show its capability in class discovery through unsupervised classification on real-life data.

By: Jayanta Basak

Published in: RI07009 in 2007

LIMITED DISTRIBUTION NOTICE:

This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.

RI07009.pdf

Questions about this service can be mailed to reports@us.ibm.com .