|
Greetings!
|
If you're looking for me resume, you can find it here. If you're a semantic web bot, my FOAF profile is here. |
|
Current
Research
My research involves using parallel algorithms for high-performance data analysis. Specifically, I have been developing an online parallel Stochastic Local Search (SLS) algorithm for finding maximal cliques in dynamic graphs, and integrating this within a parallel workflow environment which contains other state-of-the-art parallel methods, all integrated together to delivery sophisticated (near)real-time data analysis.
I am co-supervised by Alan Wagner and Holger Hoos, meaning I'm affiliated with both the Parallel Computation Lab (part of the Distributed Systems Group and the Bioinformatics, and Empirical & Theoretical Algorithmics Laboratory (ß-Lab).
The current application of my research involves finding clusters of stocks with similar intra-day activity. Using high-frequency
intra-day stock market data, we construct and maintain a dynamic graph, and employ our SLS algorithm to find cliques. The overall system
is connected using distributed components and runs in parallel on
arbitrary-sized clusters, providing the necessary computing power
to interact with the large datasets in real-time. The system is
designed to work as follows: First, we construct a pair-wise correlation matrix for a set of stocks. We then
convert this matrix into a network model, more specifically the the market graph
, where nodes represent stocks and an edge indicates that the correlation
co-efficient between two stocks exceeds a user-specified threshold. Using a state-of-the-art
SLS algorithm, we find find a set of cliques in the
market graph. Previous research has shown that, in this context, cliques
represent stocks with highly correlated price trajectories, while cliques in the complementary graph
represent a subset of stocks that are
optimally diversified over the recent time window under consideration. We have built a prototype visualization tool for monitoring and exploring the evolving intra-day market graph.
Another application of this research is in the field of bioinformatics, using gene expression
data gathered from DNA microarrays. Gene expression levels
say "how much" of a particular gene is present in a given
sequence of DNA. The approach is similar to that of the stock market application described above, except that in this case
the network model is constructed using the gene expression correlation matrix, and thus in the graph
nodes represent genes and an edge between two nodes means the genes have significantly high correlated expression levels.
In the last step, we find interesting subsets of genes by finding maximal cliques in the graph using our parallel stochastic
local search algorithm. From these computational experiments we
can answer questions like "what genes are present (or absent) in
people with disease X?" or "what genes(s) exhibited regulatory relationships?"
(allowing us to construct models of cellular signalling pathways).
Undergraduate Research
At the University of Regina, I spent two summers working with Dr. Howard Hamilton on various data-mining projects (funded by NSERC). Of particular interest was a data mining project that involved a constrained geo-spatial clustering algorithm, dubbed "DBRS: Density-based Random Sampling with Obstacles and Facilitators". You can find this paper in my list of publications.
Through undergraduate, my research interests were broad, but some of my favourite/best work was in Distributed Systems, Human-Computer Interaction and Information Systems. My honours projects were Peer-to-Peer Computing (P2P) and Sports Management Information Systems (SMIS). For more details on my undergraduate university projects click here.
Past Grad Courses
Semester 2:
- Stochastic
Local Search
Class presentations on the Traveling Salesperson Problem (TSP) can be found here
Course project was titled "Multi-level Stochastic Local Search" [paper, presentation]
- Topics
in Computer Systems
Class presentations on various subjects can be found here.
Course project was titled "R2C: A DVD Menu and Playback Control System" [paper]
Semester 1
- Algorithms
for Bioinformatics
Course project was titled "Protein Design in the 2D HP Model" [paper, slides] which was accepted to the 2005 CPSC-545 Mini-Workshop on "Algorithms in Bioinformatics".
- Artificial
Intelligence I
- Machine Learning & Data Mining (undergrad)
Publications:
- Xin Wang, Camilo Rostoker, and Howard J. Hamilton
"Density-Based Spatial Clustering in the Presence of Obstacles and Facilitators", 06/04 (ISBN 07731-0490-9)
Proceedings of ECML/PKDD-2004, the 15th European Conference on Machine Learning (ECML) and the 8th European Conference on Principles and Practice of Knowledge Discovery in Databases (PKDD). Pisa, Italy, September 20-24, 2004.
Download options: Pre-print
technical report (free), Published
edition
