Henry Tan, Setiawan BCSE (Hons)
email: henryws 'at' it.uts.edu.au
phone: +61 9514 4469
mobile: +61 415256039 (vodafone)
PhD student @ Univ. of Technology Sydney (UTS), Australia
Supervisor: Prof. Tharam S. Dillon
CV (short , long)
last updated: 24 October 2005
Recently read books:
1. Michael White, "Isaac Newton, The Last Sorcerer", Fourth Estate Limited, 1997
2. Richard P. Feynman, "The Meaning of It All: Thought of a Citizen-Scientist", Addison-Wesley, USA, 1998
3. Joseph Schwartz, "The Creative Moment: How Science Made Itself Alien to Modern Culture", Jonathan Cape, London, 1992
Tree mining has many useful applications in areas such as Bioinformatics, XML mining, Web mining, etc. In general, most of the formally represented information in these domains is a tree structured form. In this paper we focus on mining frequent embedded subtrees from databases of rooted labeled ordered subtrees. We propose a novel and unique embedding list representation that is suitable for describing embedded subtrees. This representation is completely different from the string-like or conventional adjacency list representation previously utilized for trees. We present the mathematical model of a breadth-first-search Tree Model Guided (TMG) candidate generation approach previously introduced in [6]. The key characteristic of the TMG approach is that it enumerates fewer candidates by ensuring that only valid candidates that conform to the structural aspects of the data are generated as opposed to the join approach. Our experiments with both synthetic and real-life datasets provide comparisons against one of the state-of-the-art algorithms, TreeMiner [13], and they demonstrate the effectiveness and the efficiency of the technique.
Keywords: treeminer, tree mining, frequent tree mining, mining embedded subtree, treeminer.
Citation: Henry Tan, Tharam S. Dillon, Fedja Hadzic, Elizabeth Chang, Ling Feng, MB3-Miner: mining eMBedded subTREEs using Tree Model Guided candidate generation. To appear in Mining Complex Data (MCD) '05 Workshop (as part of ICDM'05). 2005. Houston, Texas, USA.
An algorithm to mine frequent rooted ordered trees, which is best represented by XML Documents. Using Tree Model Guided candidate generation technique. The proposed technique generate less number of invalid/infrequent candidates compare to other approaches.
Citation: H. Tan, T.S. Dillon, L. Feng, E. Chang, F. Hadzic, X3-Miner: mining patterns from XML Database. in Data Mining '05. 2005. Skiathos, Greece.
There are 4 category of patterns present in tree structure: valid frequent, valid infrequent, invalid frequent and invalid infrequent. With the huge number of candidates can be generated out of arbitrary XML documents the generation of the last two categories are redundant and costly. This paper explain how the Tree Model Guided candidate generation technique reduces the number of candidates generated by avoiding the generation of invalid candidates utilizing the semantic embedded in the XML Document (tree-like structure). In this paper, a mathematical model predicting a complexity of C2 candidate generation are also described. The complexity of C2 generation is reported to be much less than O(n2).
Citation: To Appear in Data Mining 2006
Motion can be generated through physics simulation. Alternatively, an Multilayer Perceptron Neural Network based motion generation system can learn from the physics simulation to generate motion in a less expensive way but extremely accurate.
Citation: Tan, H., Rankin, J. R., Dillon, T. S., "Multilayer Perceptron as the Basis for Gaming Motion Generation." Melbourne, Australia, La Trobe University (2002).