Skip to main content
Held 8 April 2016 in 011 Sitterson Hall


3:30-3:35 Welcome

3:35-4:20 Session 1

“Mixed-focus Difficulty-Triggered Collaborative Writing: Architecture, Evaluation, and Visualization” by Duri Long, supervised by Prasun Dewan

“Alphabet Soup: Learning Knowledge Graphs from Recipes on the Web” by Adam Aji, supervised by Tamara Berg

“A Framework for Diffusion Fiber-based Analysis of T1w/T2w Ratio Map” by Haiwei Chen, supervised by Martin Styner

4:25-5:25 Session 2

“A Method for Construction of a Splice Graph from RNA Sequence Data” by Jake Bogerd, supervised by Jan Prins

“Virtual Presence for Medical Procedures” by Federico Menozzi, supervised by Henry Fuchs

“Acoustic Material Classification for Sound Propagation” by Scott Loftin, supervised by Dinesh Manocha

“Sonario” by Sarah Rust, supervised by Gary Bishop

5:30-6:15 Session 3

“Congestion Avoidance on Road Networks through Adaptive Routing on Contracted Graphs” by Bradley Davis, supervised by Diane Pozefsky

“Evaluating CUDA Applications under EDF Scheduling for use in Autonomous Vehicles” by Forrest Li, supervised by Don Smith

“Need for Speed (GPUs in Real Time Systems)” by Vance Miller, supervised by Don Smith

6:20-7:00 Reception — hors d’oeuvres will be served

Presentation Abstracts and Presenter Bios

Mixed-focus Difficulty-Triggered Collaborative Writing: Architecture, Evaluation, and Visualization

Duri Long, supervised by Prasun Dewan

Click here to watch

Many students face difficulty when writing documents due to various reasons such as language barriers, content misunderstanding, or lack of formal writing education. While some resources do exist for students who need help with their writing, many are often too shy or too busy to visit a writing center or speak with a professor during office hours. Technology also falls short in this arena. Asynchronous collaboration systems require students to self-report when they are struggling and many students tend to under-report difficulty. Synchronous collaboration systems eliminate the need for self-reporting, but require teachers to constantly monitor their students. By combining synchronous and asynchronous collaboration paradigms, this project is able to create a mixed-focus collaborative writing system in which students and teachers engage in collaboration only when triggered by an automatically generated event that indicates the student is facing difficulty

Duri Long is a senior graduating with a bachelor’s degree in computer science and a second major in dramatic art. Originally from Kansas City, she will be pursuing a doctorate in humancentered computing at Georgia Tech in the fall. In addition to working as an undergraduate teaching assistant, Duri has been involved with CS research for two years. She has worked on projects related to educational technology, collaborative systems, and data visualization.

Alphabet Soup: Learning Knowledge Graphs from Recipes on the Web

Adam Aji, supervised by Tamara Berg

Click here to watch

With the growth of the internet, there has been an increasing amount of data to learn world knowledge from; people post more pictures and write more text every day. From these, one relatively untapped source of data lies in the domain of cooking. Textual recipes are hosted on a wide range of sites, and the web is full of images of different meals and what goes into them. Furthermore, as instructional texts, recipes hold information about states of objects and what changes they undergo to reach a goal. I will present a method to try and extract this information from these data, and its use in applications such as automatically illustrating new and unseen recipes.

Adam Aji is a junior majoring in computer science and linguistics. He is interested in studying how natural language can bridge the gap between machines and people, as well as other crossdisciplinary applications in the realms of artificial intelligence, computer vision, and cognitive science.

A Framework for Diffusion Fiber-based Analysis of T1w/T2w Ratio Map

Haiwei Chen, supervised by Martin Styner

Click here to watch

The cerebral white matter in human brain develops from a mostly non-myelinated state to a nearly fully mature white matter myelination within the first few years of life. The study of myelination is of interests in a number of brain development studies. High resolution T1w/T2w ratio maps are believed to be effective in quantitatively estimating myelin content. We propose the use of a fiber-tract-based analysis of such T1w/T2w ratio data, as it is allowed to separate fiber bundles that regional analysis imprecisely groups together and associate effects to specific tracts rather than large, broad regions. We developed an intuitive tool to facilitate such fiber-based studies of T1w/T2w ratio maps. Via its GUI the tool is accessible to non-technical users. The framework uses calibrated T1w/T2w ratio maps and a prior fiber atlas as an input to generate profiles by an adapted version of the UNC atlas-based fiber analysis toolkit that is able to handle non-diffusion data. The resulting fiber profiles are used in a statistical analysis that performs along tract functional statistical analysis. We applied this approach to a study of neonate, early brain development. Results: A publicly available tool for the fiber based analysis of T1w/T2w ratio maps has been implemented and tested on a study of brain development.

Haiwei Chen is a senior at UNC majoring in computer science and sociology. He works with Dr. Martin Styner in the Neuro Image Research and Analysis Laboratories. His research interests lie in the field of computer vision. He plans to apply to a doctoral program next year to further explore the field.

A Method for Construction of a Splice Graph from RNA Sequence Data

Jake Bogerd, supervised by Jan Prins

Click here to watch

Although genetic information is stored in DNA, the RNA transcripts obtained from DNA indicate which genes are actually active in any given cell. Modern high-throughput sequencing techniques allow accurate sequencing of short RNA fragments, which may then be aligned to a reference genome. These alignments can be summarized by constructing a “splice graph”, in which nodes represent genomic coordinates and edges represent sequences that are retained or spliced out of observed transcripts. Each full-length transcript corresponds to a path through the graph. I have written software to build a splice graph from a collection of short reads aligned to a reference genome. This software incorporates variants observed relative to the reference genome as additional paths. I have also written tools to manipulate and traverse such graphs. An application of this graph is correction of noisy full-length RNA transcripts. Such a transcript may be aligned to paths through the graph in order to identify its original sequence.

Jake Bogerd is a senior from Durham, NC majoring in computer science and mathematics. He works part-time in Research Triangle Park as an intern on the analytics team at Interactive Intelligence. After graduation, he will continue this work full-time as a software engineer.

Virtual Presence for Medical Procedures

Federico Menozzi, supervised by Henry Fuchs

Click here to watch

As medical training becomes more and more complex, with students being expected to learn increasingly specialized and sophisticated procedures, the current practice of having students physically observe all procedures is becoming increasingly difficult. Some procedures are exceedingly rare, while others may rely on specialized equipment not available to the student’s institution. Additionally, some procedures can be fast-paced, and critical details might be overlooked in such a hectic environment. We present an application solution that records the procedure with multiple cameras, reconstructs the 3D environment and people frame-by-frame, then utilizes virtual reality to allow the student to walk through the reconstruction of the procedure through time. We also include several post-reconstruction enhancements, such as video playback controls, scene annotations, and introducing new 3D models into the environment. While presented in the context of medical training, our system is applicable in a wide variety of training scenarios.

Federico Menozzi is a senior computer science major. He will be heading to Google’s Franklin Street office this summer to work on Skia, Google’s 2D graphics engine used in Chrome and Android, before returning to UNC in the fall to complete a master’s degree as part of UNC’s combined B.S./M.S. program. His interests mainly revolve around the field of computer graphics and its applications.

Acoustic Material Classification for Sound Propagation

Scott Loftin, supervised by Dinesh Manocha

Click here to watch

We present a novel algorithm to generate virtual acoustic effects in captured 3D models of real-world scenes for multimodal augmented reality. Our approach leverages recent advances in 3D scene reconstruction and presents algorithms to automatically compute acoustic material properties. Our technique applies convolutional neural networks to estimate the acoustic material properties, including frequency-dependent absorption coefficients, that are used for interactive sound propagation. An iterative optimization algorithm is used to adjust the materials until a virtual acoustic simulation converges to measured acoustic samples. We have applied the sound propagation and rendering algorithm to many reconstructed real-world indoor scenes and evaluated its performance for augmented reality applications.

Scott Loftin is a senior majoring in computer science and physics. He has spent the past few months working with graduate student Carl Schissler and Dr. Dinesh Manocha on acoustic scene capture for virtual and augmented reality audio.


Sarah Rust, supervised by Gary Bishop

Click here to watch

If you can’t see, how do you know when to cross the street? This question was the motivation for developing Sonario, a 3D audio game to teach the visually impaired how to recognize sound cues in the environment. We used the Image Source Method to compute first order reverberations and implemented the game in Javascript using Web Audio API. Come to Maze Day to try it out.

Sarah Rust is a senior computer science major from Cape Girardeau, MO. She spent last summer interning at Apple Inc. on the Platform Architecture Graphics team (team iPhone) and is returning again this summer, despite the FBI’s best efforts. Sarah has been working with Dr. Gary Bishop this year on developing a 3D audio game for the visually impaired, drawing from her previous research with the GAMMA group on realistic sound simulation. When Sarah asked Gary what she should write for her bio, he said, “Sarah is awesome.”

Congestion Avoidance on Road Networks through Adaptive Routing on Contracted Graphs

Bradley Davis, supervised by Diane Pozefsky

Click here to watch

We have developed a method of integrating live traffic information into preprocessed graphs of large road networks in order to adaptively route autonomous vehicles. Our intent is to reduce congestion caused by fleets of centrally-routed vehicles being assigned overlapping routes and to help those vehicles avoid already congested areas. Recent developments in shortest-path routing, namely Contraction Hierarchies, are used in conjunction with a modified bidirectional Dijkstra search algorithm to ensure fast route computations despite frequent graph updates. We introduce a novel heuristic for graph reprocessing that enables quick updates alongside a simple approach to computing appropriate edge weights based on substantial amounts of feedback received from vehicles on the road. Our approach is tested on a developed simulation platform using real road data and a Nagel–Schreckenberg traffic model. Early results show that vehicles experience an overall speedup in travel time and adeptly react to unforeseen conditions by using alternative routes to avoid further congestion.

Bradley Davis is a senior majoring in computer science. During his time at Carolina, he has been actively involved on campus, studied abroad at University College London, and participated in many programming contests, hackathons, and case competitions. He has most recently joined a local startup, LineLeader, as a principal engineer. After graduation, Bradley will move to San Francisco to work for the popular business chat application Slack as a software engineer. Bradley hopes to pursue a master’s degree in computer science through Georgia Tech’s OMSCS program in the near future.

Evaluating CUDA Applications under EDF Scheduling for use in Autonomous Vehicles

Forrest Li, supervised by Don Smith

Click here to watch

Autonomous driving vehicles and Advanced Driver Assistance Systems(ADAS) require real-time processing of sensor data to implement safety-critical features such as pedestrian detection, lane tracking, and obstacle avoidance. In vehicles with cameras, image processing applications must be able to consume and analyze image frames in real-time to be able to process a vehicle’s environment and provide feedback to either a human driver or a computer control system. CUDA, a programmable GPU platform, provides a fast parallelized image processing platform, but does not guarantee real-time deadlines are met when multiple CUDA applications are running. In this study, I evaluate CUDA applications running under the Earliest Deadline First (EDF) scheduler, which ensures that all jobs complete by their deadline. This is carried out on a Nvidia Jetson TK1, a GPU-capable board portable enough to be utilized as an image processing component within a vehicle.

Forrest Li is a senior computer science major at UNC. He currently works with Dr. Don Smith as part of a GM contract. His interests include operating systems, security, and kernel-hardware interfaces.

Need for Speed (GPUs in Real Time Systems)

Vance Miller, supervised by Don Smith

Click here to watch

In recent years, advances in computing power have reached the automotive industry in the form of advanced driver assistance systems (ADAS). ADAS takes the form of adaptive cruise control, automatic brakes, lane departure warning, collision avoidance, and even driver drowsiness detection. Cars that support these features have a variety of sensors, including cameras, that gather data from their surroundings. This data must be processed and the results acted upon in real time to prevent catastrophic failure and loss of life. GPUs offer a large speedup for image processing but their runtime performance can be unpredictable in some cases. In my research, I examine the characteristics of GPUs that make them (un)suitable for real time tasks and explore mechanisms for making them more predictable.

Vance Miller is a senior computer science major from Ocean, NC. He has spent the past year working in the Real-Time Systems group on “the car project” with Drs. Don Smith and Jim Anderson but also enjoys breaking into servers and safes with Dr. Fabian Monrose between classwork. After graduating, he plans to visit his 22-pound cat and then attend to pursue a doctorate in computer science.