Skip to main content
Held 28 April 2010 in 011 Sitterson Hall

Virtual Percussion Instruments with Touch-Enabled Interfaces

Maggie Zhou, supervised by Ming Lin

We present a virtual percussive musical instrument system that can be used to emulate playing percussive music in the real world. A touch-enabled interface including multi-touch display is used as the input device for playing the virtual instruments. The digital music is synthesized using our real-time physics-based sound synthesis engine that creates musical tones based on the user’s actual interaction with the virtual instrument. Such a system offers users the ease of performing digital percussive music in an intuitive fashion. In addition to multi-touch display, both collaboration among multiple performers and styluses for bimanual operation are also supported by our system. We demonstrate our multi-modal, physically-inspired interfaces on playing

Maggie Zhou is a junior from Cary, N.C., majoring in computer science and minoring in comparative literature. She has spent (barely) more of her life in Canada than in North Carolina, and as of the writing of this bio, her 2010 NHL bracket has yet to be broken.

What if …? Scaling Network Traffic For Realistic Experimentation

Shaddi Hasan, supervised by Kevin Jeffay

Networking researchers perform controlled experiments to test new protocols and techniques they develop. A critical variable for the network traffic commonly used as input for such experiments, known as a trace, is its offered load — how much data is transmitted per unit time. Modifying a trace’s offered load is a desirable ability for performing controlled experimentation, but performing such modification while preserving important unrelated characteristics of the trace is difficult. I examine the effects of one technique for achieving this goal, block resampling, and propose modifications for adjusting the offered load of a trace while preserving its other fundamental characteristics. I show that the existing block resampling technique, despite achieving its original design goals, can produce unpredictable result traces and introduces biases into result traces that effect important protocol performance measures such as queue lengths and number of active connections. I then provide modifications to the basic block resampling algorithm that more faithfully preserve these characteristics in the way they were expressed in the original input trace.

Shaddi Hasan is an aspiring Internet plumber. He came to Carolina frustrated with the effort he had personally put into unclogging the network tubes of his friends, neighbors, and family, and once he arrived he learned that he was not alone in his frustration: indeed, the clogs he had seen at home in Knoxville, Tenn., paled in comparison to some of the issues that professional Internet plumbers faced on a daily basis. The thought of unclogging an Internet tube filled with millions of kitten videos terrified him, as did the thought that within a few years those millions would be billions. Since that time, he has worked under the guidance of Dr. Kevin Jeffay to develop tools and techniques to better understand how kittens fit inside the Internet tubes, work which hopefully will enable the world’s Internet plumbers to push more kittens through those tubes than ever before.
In the fall, Shaddi will be jumping down the warp tube to begin graduate study at Berkeley. In the coming years he hopes to work on unclogging tubes in the developing world and to help educate the next generation of tube plumbers and researchers.

SPACE: Graph Classification Based on Discriminative Spatial Clusters

Calvin Young, supervised by Wei Wang

Graph classification is an increasingly important step in applications such as drug design, function prediction of molecules, and automated analysis of program work flow. Among the various techniques proposed in the literature, graph classification based on discriminative subgraph patterns is a popular strategy; however, these methods are often hampered by the computational complexity of mining subgraph patterns in large datasets and by the lack of global discrimination power of individual patterns.
An interesting observation in pattern mining is that discriminative subgraphs tend to form naturally occurring spatial clusters, particularly in graphs representing organic compounds where these clusters might represent structures such as active site features. We exploit this observation and study the feasibility of building graph classifiers from discriminative spatial clusters. We propose an algorithm that first employs a pattern mining algorithm to identify potential hot spots and then iteratively focuses mining efforts in these regions. In this manner, the algorithm incrementally refines spatial clusters until they converge. This algorithm can use any subgraph pattern mining algorithm, but our implementation employs an algorithm COM we previously developed. Not only does COM outperform many existing pattern mining algorithms in terms of classification accuracy and time efficiency, it also mines for pattern co-occurrences, which aids in the formation of spatial pattern clusters.
Experiments show promising accuracy improvements over other state-of-the-art methods. Furthermore, graph classifiers built on discriminative spatial clusters show the greatest increase in classification accuracy when used with moderately sparse graphs.

Calvin Young is a senior computer science and business administration double major from Apex, North Carolina. Calvin has worked under the mentorship of Dr. Wei Wang for the last two and a half years. In his spare time, Calvin enjoys playing table tennis, violin, and singing karaoke. This past summer, Calvin interned at Google, Inc. in Mountain View, Calif., where he will return after graduation as a software engineer.