My work is centered around data mining, the science of discovering latent and useful knowledge in large databases (also known as KDD - Knowledge Discovery and Data mining). As part of the group (Data Mining Template Library - DMTL, see Projects below) under Prof. Zaki, I am currently working on frequent pattern mining (FPM). FPM is a category of problems in data mining that deals with finding frequent patterns (viz. itemsets, sequences, trees or graphs) in massive relational databases. Unfortunately FPM (even data mining for that matter) is not well documented - largely because data mining is a relatively new field which is witnessing rapid new developments. Some useful pointers on data mining:

• IBM Quest project whitepaper
• www.kdnuggets.com
• Statistical Data Mining Tutorials
• More tutorials on data mining
• A link to upcoming DM conferences
  
  

Frequent patterns aren't always "interesting", but that would be the focus of another problem. Our group is working towards developing a unified and generic framework for FPM. We currently have efficient and functional modules for itemsets, sequences, trees and graphs as part of our library. My work in the group has largely revolved around tree and graph mining (see references below for relevant papers). Graph mining is particularly challenging because of the inherent complexity of graphs, raising issues like graph isomorphism (a problem known to be NP, but not proved to be NP-complete). We aim to have an open source release of our library soon - watch this space!

References - (please also check the publications section)

• Technical Report of our current work
• gSpan (a graph mining algorithm)
• TreeMiner (Prof. Zaki's tree mining algorithm, also look up the errata here)
• Mining Structured Data

Projects

• Data Mining Template Library: Under progress
  Developing a generic and unified framework for efficient frequent pattern mining; comprising itemset, sequence, tree and graph mining components.
  Resources used: C++ (using STL)
  DMTL's public release
  
  
• DNA Sequence Inhomogeneity using log-odds (course project in Algorithms in Computational Molecular Biology): April 2004
  Design and implementation of a scheme to detect inhomogeneity in long DNA sequences using log-odds analysis.
  Resources used: C++
  
  
• Network Service Design (course project in Network Programming): April 2004
  Conceptual design of a scalable, reliable and secure Instant Messaging System (ala MSN). Read my design document here.
  
  
• Simulator of Memory Management Unit (course project in Operating Systems): October 2003
  Development of a complete emulator of the Memory Management Unit (MMU) of an operating system.
  Resources used: C
  
  
• Characterization of Video Sequences (Major BS Thesis): December 2002 - May 2003
  The project applies artificial intelligence concepts for information retrieval from video   sequences. The video clips are first segmented, following which information pertaining to the clips is extracted. This information is analyzed using an intelligent system to categorize the sequences into distinct classes.
  Resources used: Matlab
  
  
• Study of Classification using Neural Networks and Data Mining Concepts: June - July 2002
  Study of classification concepts, among them prominent classification tools like neural networks and decision trees. A comparative analysis was made on the performance of these two techniques. A popular data mining algorithm, SLIQ was used for classification by decision trees, along with back propagation algorithm for neural networks.
  Resources used: Matlab, C++
  
  
• Association Rule Mining: Extension of Direct Hashing & Pruning Concepts to Quantitative Databases: March 2002
  The project involved application of the Apriori algorithm for association rule mining, with extension using Direct Hashing and Pruning (DHP) concepts to quantitative databases.
  Resources used: Java, Oracle
  
  
• Handwritten Character Recognition using Neuro-Fuzzy Techniques: July - September 2001
  Devising a Neuro – Fuzzy technique for character recognition, combining the strengths of Fuzzy Logic to overcome the drawbacks in traditional character recognition systems that employ Artificial Neural Networks alone.
  Resources used: Matlab