September 25, 2009
Students discussed their summer research/work experiences.
Dan Fast “Comedy Central Game Content Management”
Erdem Sahin “Networks Research”
Mayada Gonimah “My Work at Goldman Sachs”
Andrew Lorenzen “Research in Economics”
Katie Creel “Robotics”
October 2, 2009
Prof. Stephen Freund, Williams College
FastTrack: Efficient and Precise Dynamic Race Detection
Multithreaded programs are notoriously prone to race conditions. Prior work on dynamic race detectors includes fast but imprecise race detectors that report false alarms, as well as slow but precise race detectors that never report false alarms. The latter typically use expensive vector clock operations that require time linear in the number of program threads.
Our new approach to race detection exploits the insight that the full generality of vector clocks is unnecessary in most cases. That is, we can replace heavyweight vector clocks with an adaptive lightweight representation that, for almost all operations of the target program, requires only constant space and supports constant-time operations. This representation change significantly improves time and space performance, with no loss in precision.
Experimental results on Java benchmarks including the Eclipse development environment show that our FastTrack race detector is an order of magnitude faster than a traditional vector-clock race detector, and roughly twice as fast as the high-performance DJIT+ algorithm. FastTrack is even comparable in speed to Eraser on our Java benchmarks, while never reporting false alarms.
This is joint work with Cormac Flanagan (University of California, Santa Cruz).
October 9, 2009
Computer Science Faculty discussed Graduate school - how to apply and what to expect.
October 23, 2009
Speaker: Douglas Turnbull, Ph.D.
Swarthmore College
Combining Audio Content and Social Context to Improve Music Discovery
Most commercial music discovery engines (e.g., Apple iTunes Genius, Pandora Radio, Last.fm) rely on context-based analysis of social information (e.g., user preferences, blogs, or social tagging data) to find music. These systems are not ideal in that they suffer from popularity bias and the “cold start” problem. To remedy these problems, researchers have been exploring content-based audio analysis as an alternative. However, the state-of-the-art content-based systems can sometimes produce less-than-accurate annotations of music.
It seems natural that we can improve music search by combining acoustic and social sources of music information. In this talk, I first described how we can collect social information about music and how we can automatically annotate music by analyzing audio signals. I then compare three different approaches (calibrated score averaging, RankBoost, and kernel combination SVMs) that improve music search by combining audio content with social context.
Lastly, I showed a demo of Meerkat, a web-based semantic music discovery engine, that works like a personalized Internet radio player, but allows a user to control the stream of music using terms like "bluegrass", "acoustic instrumentation", and "melodramatic". Meerkat was originally conceived as a collaborative class project in an undergraduate course on information retrieval. My students hope to launch a public version in the coming months.
Bio:
Douglas Turnbull is currently a visiting assistant professor in the computer science department at Swarthmore College. His main research interests include multimedia information retrieval, computer audition, machine learning, and human computation.
In 2008, Doug graduated with a Ph.D. in computer science from UC San Diego where he was an NSF IGERT fellow. While at UCSD, he co-founded the interdisciplinary Computer Audition Laboratory (CAL). In his first year at Swarthmore, he submitted six manuscripts with eight undergraduate authors to four premier international conferences.
November 13, 2009
Computer Science Alumni Panel discussed life after Williams.
November 20, 2009
Evaluating Natural Language Generation Systems in Online Virtual Environments
Kristina Striegnitz, Ph.D.
Union College
The GIVE challenge (GIVE = Giving Instructions in Virtual Environments) is a recent shared task in which natural language generation systems are evaluated over the Internet. In the GIVE scenario, human users enter an online 3D virtual environment where they need to solve a treasure hunt with the help of automatically generated instructions. The challenge for the natural language generation systems is to produce, in real time, effective English instructions that guide the users to their goal.
In this talk, Kristina explained the rationale for the GIVE challenge, described the software infrastructure that was developed to support it, and presented the results of the first evaluation round and of a parallel laboratory experiment. The data collected over the Internet are consistent with the data collected in the more traditional laboratory setting, but the Internet approach offers the statistical power for more fine-grained evaluations because it is cheaper to collect large amounts of data.
For more information about the GIVE challenge see www.give-challenge.org/research.
=======================================
Kristina Striegnitz is an assistant professor at Union College. Before coming to Union, she received a joint PhD from Saarland University in Germany and the University Henri Poincare in France, and did a Postdoc at Northwestern University. Her research is in natural language generation, dialog systems, and embodied conversational agents.
