590AI Spring '01

Graduate Research in Artificial Intelligence

Greetings, and welcome to 590AI, Spring 2000! The focus of this seminar is research being done here at UW by our very own graduate students, as well as the occasional professor and guest lecturer crashing the party.  You may also volunteer to give a talk about work by other people that you find fascinating and relevent.   If you're interested in AI, then be sure to sign up for this seminar!

Administrivia

Like most all other classes in the department, CSE 590AI has a mailing list, managed by majordomo. Please sign up by sending a message to majordomo@cs.washington.edu with the text subscribe cse590ai in the body. You can view an archive of the mailing list online as well.

Meeting time: Wednesday 4:30 - 5:20
Meeting place: EE1 003
Organizer: Henry Kautz <kautz@cs.washington.edu>
 

About credits

590AI is offered as a 1 - 3 credit course. Signing up for one credit is, of course, the minimum, and the minimum will be expected of you: attend the seminars, listen to the presentation, and participate in discussions. If you'd just like to get a feel for AI research here at the UW, then sign up for one credit of 590AI and enjoy the presentations.

If, on the other hand, you have an active interest in AI and would like to present during the quarter, then you should sign up for 3 credits of 590AI. We're asking folks to present either their current research work or a recent paper of common interest (a list of possible papers is below, but feel free to suggest a paper yourself). Don't worry, we're not asking for AAAI conference talks, here; we're more casual than that. What we're after is a talk that is accessible to folks who have read (or at least skimmed) the paper you're presenting, but who may not have read any other related work. If you're presenting your own work, it'd be great if you could include results / findings / whatever that has you excited about your work this week.

About cookies

Each week's seminar will be catered by one of the students in the class. The "Cooking supplier" should bring enough treats (typically cookies, but feel free to be creative) for all. Fare from the past has ranged from Peppridge Farms cookies bought at the HUB newsstand, all the way to peanut brittle, brownies, and Kool-aid, all prepared by hand. If you're worried about being adequate, just don't follow Steve Wolfman.

The schedule

 
Wednesday Presentation Cookie supplier
3/28 Corin Anderson, "Adaptive Web Navigation for Wireless Devices" Henry Kautz
4/4 Anhai Doan, "Recounciling Schemas of Disparate Data Sources: A Machine Learning Approach"

Given a source and target schemas, the schema matching problem is to find semantic mappings between the elements of the two schemas, such as "address" maps to "location", and "num-baths" maps into "half-baths" + "full-baths". Schema matching is a critical step in numerous data management applications. As such, it has motivated many solution proposals. Recenly, the proposals have focused on developing generic schema matching systems that are robust and applicable across domains.

In this talk I describe the LSD project which uses and extends machine learning techniques to build a generic schema matching system. I begin by discussing the challenges and requirements for such a system. Next, I describe how LSD meet these challenges and outline the system architecture. Finally, I describe the work we have done in the context of LSD and discuss future research.

Dave
4/11 Don Patterson, "Improving donor splice site recognition in human pre-mRNA using predicted structural cues."

In this paper we derive six metrics from predicted secondary structure of human pre-mRNA that can be used to recognize donor splice sites. We compare the performance of a decision tree ensemble that is trained on these metrics to a primary sequence weight matrix discriminator(Simple Bayesian classifier) and demonstrate that secondary structure complements weight matrix techniques and improves overall classification rates. Finally, we show that structure based techniques improve discriminating power in a subclass of genes for which a primary sequence based method is guaranteed to fail. 

Sarah
4/18 Jim Brinkley , Cornelius Rosse rosse@biostr.washington.edu>

 Current and Potential AI and Other CSE Research Problems in the UW Structural Informatics Group

The UW Structural Informatics Group is an interdisciplinary research program located in the UW Health Science complex. Led by Jim Brinkley (who has an adjunct faculty appointment in CSE) and Cornelius Rosse, with significant collaboration with Linda Shapiro in CSE, the goals of the group are to 1) develop methods for representing, organizing, managing and utilizing information about the physical organization of the body, and 2) use these methods for solving real-world problems in clinical medicine, education and research. The overriding hypothesis is that structure is the foundation for understanding in biomedicine. If we can develop methods for representing and organizing structure, then we will be in a position to organize much of the data and knowledge in biomedicine. The research is application-driven, leading to many interesting CSE problems. Current applications of interest, all of which are aimed at delivery over the Internet, are anatomy education, radiation treatment planning, and brain mapping. CSE problems that arise out of these applications include multimedia and distributed database management, computer graphics and visualization, image understanding, and symbolic knowledge representation and reasoning. The latter problem has led to the development of a Foundational Model of Anatomy (FM), which is implemented in the Stanford frame-based knowledge acquisition tool called Protege. By means of a foundational model server (FMS) the FM is accessed by various client applications for intelligent 3-D scene building and anatomic query processing. These and other potential applications and research problems will be described and demonstrated. 

Julie
4/25 David Hsu, "Silicon Learning" Yongshao
5/2 Geoff Hulten Deepak Verma
5/9 Corin Anderson, "Report on the Hong Kong WWW Conference" Anhai Doan
5/16 Matt Richardson Geoff Hulten
5/23 Sarah Schwarm, "Speech Recognition" Matt Richardson
5/30 Yongshao Ruan, "Learning Restart Policies"  Don P.


Send corrections to
kautz@cs.washington.edu