Thursday, October 4, 2012

GHC12: SRC Poster Competition: Graduate - Round 2

In this session semifinalist graduate students from the Wednesday night poster session are each given 15 minutes to present their work.

Jaya Kawale, University of Minnesota, presented on teleconnetions in climate data. I got to learn about dipoles that represented different classes of teleconnections characterized by anomalies. These are important, as they can cause temperature and precipitation anomalies throughout the globe. These anomalies can lead to droughts or monsoons, being able to predict them can help people better prepare for severe weather.

Kawale's research found that these weather anomalies had a partner event, so her research worked on correlating that and clustering the data.  By using automatic dipole discovery, most dipoles have been discovered and new ones are found quickly.  This data can also help with quantifying impact on land temperature.

Awalin Sopan, University of Maryland , started her presentation with a story about a series of crimes in a community. Where the police did not see the pattern, neighbors communicating online discovered the pattern and the crime was solved. This became Nation of Neighbors, a very successful way for neighbors to track crime in their own community.  Her team worked with them, along with social scientists, to find out what constitutes a successful community, classified by their tool ManyNets.  The tool itself is generic - it analyzes networks.

All communities have a lot of invitations, but successful ones had more reports. People tended to want to report things anonymously - but then how can you identify the top influencers?  They did it by looking at just basic activity.

More leaders in a community resulted in a more active and more beneficial community. Interestingly, very few of the leaders had anything to do with law enforcement, just concerned citizens that took on this role voluntarily.

Unfortunately, the researchers were unable to determine if the community became more safe or not after participation (no access to the data). Though, by participating, neighbors felt more safe.

Zalia Shams, Virginia Tech, presented her research on cross-testing. A pretty ingenious concept - when a CS class has an assignement, the TA needs to submit test programs to run against the programs, but then they also had the students submit their own test cases. Of course, most students will pass their own test cases, but then they would run other student's tests against another student's code - a great way to find bugs! (personal note: I always think that tests are better if someone else writes them. When you write your own tests, you'll miss your blind spots).

This isn't as simple as it sounds. Depending on the program, the test cases may need to be built against the submission, which lead to lots of compile errors. The researcher used late binding to work around this.

Just like in the real world, students weren't very excited about writing lots of test cases for their own project, but did like the idea of their tests cases being used against other students programs.  In one of their tests, they found that 0 submissions passed 100% of the tests. They had removed tests that were testing things that were outside of the scope of the assignment.

This is like crowd sourcing your tests. A fascinating idea, but unlikely to work outside of a university setting where you're forced to write many test cases :-)  But, I think this is a great idea to teach robust programming, and I do wonder how we could apply this to the industry.

This was all based on extending Web-CAT.

This post syndicated from Thoughts on security, beer, theater and biking!

No comments:

Post a Comment