Professor Johannes Gehrke in Computer Science works in database systems, data mining, and data privacy, with a focus on applied research that has strong conceptual foundations. His group is building an event-monitoring system that enables users to monitor distributed XML streams such as RSS feeds or events from RFID readers. The novelty of the system, called Cayuga, is that it allows users to register complex stateful queries. A stateful query can involve several events over time as in the following example: “Once www.wired.com posts an article with data in its title, send me the first new posting from www.slashdot.org with a link to this article in its summary text.”

Challenges include the design and implementation of Cayuga such that the system can scale to very high event throughputs while processing thousands of concurrently registered queries.

In other research, Gehrke’s group is collaborating with companies such as Capital One to apply data-mining techniques to problems in finance and marketing, and he is working with scientists to apply data mining to scientific problems. For example, he is working jointly with astronomers from the Arecibo radio telescope observatory on a new census of all pulsars in the Milky Way galaxy. His group has developed some of the fastest data-mining algorithms available today, and his current focus is to improve the quality and scalability of data-mining methods.

His third research direction is data privacy. His group is working on practical methods to share and analyze data while not disclosing private information in the data.

Professor Thorsten Joachims in Computer Science is developing search engines that go beyond today’s one-size-fits-all paradigm. His goal is to make search engines adapt to a user and a document collection, giving them the ability to learn from mistakes. He and his colleagues showed that observing how users interact with the system (for example, clicks and query reformulations) can provide valuable feedback about how satisfied the user was with the results of a query. Joachims is developing machine learning methods that can use this feedback to adapt the search engine’s ranking function so that it serves the user better in the future.