Archive for September, 2008

Anomaly Pattern Detection in Categorical Datasets

Day 3 started with Invited Talk of Dr. Michael Schwarz from Yahoo! Research on “Internet Advertising and Optimal Auction Design” who discussed Generalized English Auctions  and Internet advertising, generalized second price options. (Details can be read on Akshay Java’s Social Media research blog here). It was an interesting talk pertaining to how almost every transaction follows the modern auction model and what approaches can be used to maximize the throughput and ROI.
Later during regular sessions, I attended the Discovery and Detection research session which was focused on outlier analysis. It comprised of the following presentations.
25-minute presentations

• Automatic Identification of Quasi-Experimental Designs for Discovering Causal Knowledge. D. D. Jensen, A. S. Fast, B. J. Taylor, M. E. Maier.
• Discrimination-aware Data Mining. D. Pedreschi, S. Ruggieri, F. Turini.

15-minute presentations

• Local Peculiarity Factor and Its Application in Outlier Detection. J. Yang, N. Zhong, Y. Yao, J. Wang.
• Angle-Based Outlier Detection in High-dimensional Data. H. Kriegel, M. Schubert, A. Zimek.
• Anomaly Pattern Detection in Categorical Datasets. K. Das, J. Schneider, D. B. Neill.

Lunch was sponsored by Yahoo! and it was pretty cool décor with their gadgets, puzzles and YahooDokus.  Later in the afternoon I attended
25-minute presentations

• Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering Model. Y. Koren.
• Combinational Collaborative Filtering for Personalized Community Recommendation. W. Chen, D. Zhang, E. Y. Chang.

And then

• (R) Scaling Up Text Classification for Large File Systems. G. Forman, S. Rajaram.
• (I) Data Mining Using High Performance Data Clouds: Experimental Studies Using Sector and Sphere. R. Grossman, Y. Gu Short.

I later got a chance to talk to Dr. Grossman about the cloud computing initiative that he is very passionate about. He discussed in his presentation how Sector is approximately twice as fast as Hadoop and how Sector has been used to distribute the Sloan Digital Sky Survey (SDSS) via the web site sdss.ncdm.uic.edu.
The day concluded with Poster reception where I get to talk to several authors and presenters including Wen-Yen Chen, Pooja Mittal, Yabo-Arber Xu, Kaustav Das of Anomaly Pattern Detection in Categorical Datasets and one of the authors of Information Extraction from Wikipedia, not sure who.

Day 4

The last day of conference started with Jitendra Malik’s invited talk on “The Future of Image Search”.  (Greg Linden’s Blog Post about the talk). It was a great talk where Jitendra discussed the evolution of vision, image search, shortcomings of tagging and textual taxonomies and pushed for "category recognition" for objects in images.
Later there were the following excellent sessions.

• Context-Aware Query Suggestion by Mining Click-Through and Session Data. H. Cao, D. Jiang, J. Pei, Q. He, Z. Liao, E. Chen, H. Li. (Best Application Paper Award Winner)
• Scalable and Near Real-Time Burst Detection from eCommerce Queries. N. Parikh, N. Sundaresan.
• Using Predictive Analysis to Improve Invoice-to-Cash Collection. S. Zeng, P. Melville, C. A. Lang, I. Boier-Martin, C. Murphy.
• TagMark: Reliable Estimations of RFID Tags for Business Processes. L. W. F. Chaves, E. Buchmann, K. Böhm.

The conference concluded with closing remarks from the general chair, Ying Li.

Day 2 started with Trever Hastie’s talk on regularization paths and coordinate descent.  It was great to see Dr. Hastie passionately speaking about the coordinate descent, logistic regression and fitting.  The keynote’s topic was “Regularization Paths and Coordinate Descent” and following is a brief abstract from the talk.

“In a statistical world faced with an explosion of data, regularization has become an important ingredient. In many problems, we have many more variables than observations, and the lasso penalty and its hybrids have become increasingly useful. This talk presents some effective algorithms based on coordinate descent for fitting large scale regularization paths for a variety of problems. Joint work with Rob Tibshirani and Jerome Friedman”

After the keynote talk, there were combined and research sessions. I attended one with social Networks which comprised of the following presentations.
25-minute presentations

• The Structure of Information Pathways in a Social Communication Network. G. Kossinets, J. Kleinberg, D. Watts.
• Influence and Correlation in Social Networks. A. Anagnostopoulos, R. Kumar, M. Mahdian.
• Weighted Graphs and Disconnected Components. M. McGlohon, L. Akoglu, C. Faloutsos.

15-minute presentations

• Microscopic Evolution of Social Networks. J. Leskovec, L. Backstrom, R. Kumar, A. Tomkins.
• Mobile Call Graphs: Beyond Power-Law and Lognormal Distributions. M. Seshadri, S. Machiraju, A. Sridharan, J. Bolot, C. Faloutsos, J. Leskovec.
• Feedback Effects between Similarity and Social Influence in Online Communities. D. Crandall, D. Cosley, D. Huttenlocher, J. Kleinberg, S. Suri.

During lunch which was sponsored by Microsoft adCenter labs, they talked about challenges in advertising and applying it to get the best out of revenue share and context base hits. They also announced adCenter labs challenge which I have yet to find any information about online.
Later in the afternoon, Microsoft Research’s Thore Graepel talked about Large Scale Data Analysis and Modeling in Online Services and Advertising. It was a very interesting and pragmatic presentation about the real world problems in online search and advertising. Even though the first part of presentation was online ranking and matchmaking heavy, the later discussion on advertising made up for it.
Then there were 25-minute presentations

• ArnetMiner: Extraction and Mining of Academic Social Networks. J. Tang, J. Zhang, L. Yao, J. Li, L. Zhang, Z. Su.
• Identifying Authoritative Actors in Question-Answering Forums – The Case of Yahoo! Answers. M. Bouguessa, B. Dumoulin, S. Wang. (the paper was on Belinko’s website but not there anymore, interesting?)

And on a separate track,
15-minute presentations

• Automated Cyclone Discovery and Tracking using Knowledge Sharing in Multiple Heterogeneous Satellite Data. S.-S. Ho, A. Talukder.
• Spotting Out Emerging Artists Using Geo-Aware Analysis of P2P Query Strings. N. Koenigstein, Y. Shavitt, T. Tankel.
• Land Cover Change Detection: A Case Study. S. Boriah, V. Kumar, M. Steinbach, C. Potter, S. Klooster.

Later after the automated cyclone discovery session, I got a chance to meet Dr. Talukder of Jet Propulsion Laboratory/NASA to talk about a mutual acquaintance, Dr. Homayoun Seraji

This concluded day 2 of the conference.