8th Undergraduate Research Symposium (2014)
Held 25 April 2014 in 011 Sitterson Hall
Analysis and QSAR Modeling of Human Intestinal Transporter Database
T.J. Tkacik, supervised by Diane Pozefsky
Membrane transport proteins are the molecular gatekeepers that regulate the movement of chemicals into and out of every cell of every living organism. In this study, a cheminformatics approach was taken to predict the substrate and inhibitory activities of 14 major human intestinal transporters using quantitative structure-activity relationship (QSAR) models built from 56 datasets. Dataset compounds were represented using two types of chemical descriptors and modeled using three supervised learning techniques. The predictive power of these predictors was analyzed for correlations with characterizing data of the original datasets.
T.J. Tkacik is a senior majoring in computer science and chemistry from Fort Mill, South Carolina. He is especially interested in algorithms and the application of computational methods to research in the natural sciences. After graduation, T.J. will work as a software developer for Epic Systems before likely completing graduate education in computer science or cheminformatics.
NoSQL, Neo4j, and Chemotext: A Primer on Picking the Right Database For Your Problem Space
Brian Gottfried, supervised by Diane Pozefsky
NoSQL databases get a lot of media attention nowadays, but are they anything more than a flash in the pan? Relational databases have been around for several decades and are used by almost every major business in the world, while most NoSQL databases could still be considered beta software. This case study examines a specific database, Neo4j, and a specific problem, the Chemotext application, to understand why a Graph Database like Neo4j was not only the best database for the application, but the right solution to the problem. This paper also examines Graph Databases on a larger scale to determine what problem spaces they thrive in and where they fail.
Brian Gottfried is a native of Westlake, Ohio and a happy transplant to the sunny South. His areas of interest include machine learning, data analytics, and big data manipulation. After graduation, he’ll be heading to Madison, Wisconsin to work with Epic Systems doing Software Development. He’d like to thank Ian Kim, Nancy Baker, and Diane Pozefsky for their support during his research; he is also, without a doubt, proud to be a Tar heel.
Implementing Mobile Health Records in Low-Resource Settings
Gautam Sanka, supervised by Hye-Chung Kum
The gradual adoption of electronic medical records (EMR) in resource-rich nations has improved the quality of health care and reduced medical errors. The same can be employed in resource-poor nations, and with the recent advances in mobile technology and wireless infrastructure, EMRs will become an essential part of the medical process. This paper provides guidelines on how to develop and deploy the proposed system in low-resource settings. A case study details how the proposed model was then successfully implemented at the CARE Rural Health Mission in India.
Gautam Sanka is a graduating senior from Morrisville, NC who will be pursuing a Master of Science in computer science from UNC next year. Gautam is interested in the fields of data mining, networking, and mobile computing.
Autonomous Navigation for Micro-Air Vehicles Using Reciprocal Velocity Obstacles
Hannah Rae Kerner, supervised by Dinesh Manocha
Recent developments in autonomous flying vehicles, from industry drones to quadrotors, have generated considerable interest in the development of autonomous navigation and control techniques. Our goal is to develop automatic collision-avoidance algorithms that can account for physical constraints for micro-air vehicles (MAVs) in three dimensions. Our work builds on earlier work on collision avoidance based on reciprocal velocity obstacles (RVOs) for robots in two dimensions. We extend this work to handle kinematic and dynamic constraints for MAVs. We use the ArduCopter quadrotor helicopter (quadcopter) by 3D Robotics as the underlying physical agent along with MAVLink, a communications protocol for micro-air vehicles, to interface between the quadcopters and the RVO software library. We present preliminary results from our simulator and hardware integration.
Hannah Rae Kerner, originally from Charlotte, North Carolina, is completing her Bachelor of Science in computer science in three years at UNC. She will return to UNC next year to complete her Master of Science. She has interned for NASA for three years at both Langley Research Center and Goddard Space Flight Center, working on projects including the design of the Unmanned Aircraft System Airspace Operations Challenge, a NASA Centennial Challenge. This summer, she will intern for Planet Labs, an agile aerospace startup in San Francisco, working on ground station software and satellite systems. Hannah’s technical interests are air- and spacecraft systems, machine learning, and wearable computing. Her ultimate goal is to become an astronaut.
User Interface and Design Optimization for Concentric Tube Robots
Cenk Baykal, supervised by Ron Alterovitz
Concentric tube robots have the potential to enable novel, minimally invasive surgical procedures. These medical robots are capable of traveling curvilinear paths, maneuvering around anatomical obstacles, reaching targets in tight spaces. Concentric tube robots are composed of customizable, pre-curved tubes that the user can independently rotate and extend to achieve tentacle-like motion. The pre-curvatures and lengths of the tubes have a significant impact on the robot’s maneuverability and effectiveness in performing a surgical task. Moreover, operating these robots is hard due to their complex and unintuitive kinematics. In this presentation, we demonstrate a user interface that utilizes a novel motion planning method to compute collision-free motion plans for the concentric tube robot. The user interface allows the surgeon to continuously and freely drive the robot’s tip via a 3D input device while the planner ensures the robot’s shaft does not collide with any anatomical obstacles. Furthermore, we present our preliminary work on the optimization of the robot’s design on a patient- and application-specific basis.
Cenk Baykal is a junior undergraduate student majoring in computer science and mathematics. He is fascinated by multidisciplinary research fields in computer science, such as robotics and artificial intelligence. He is especially enthusiastic about mathematically-oriented research in motion planning and optimization. After his undergraduate education, Cenk hopes to pursue graduate studies in computer science.
Audio Signal Representations For Sound Design
James Norton, supervised by Ming Lin
Despite sophisticated models of representing sound, modern software synthesizers rely heavily on analog-modeled hardware in their design. This leads to proliferation of control parameters which is evidenced by the user interface of many synthesizers. By representing audio as a path through the space of power spectral density functions, control parameters can more directly relate to perceptually important features. This creates a more streamlined process, allowing users to focus more on the sound than on the signal flow used by the implementation.
James Norton is a Chapel Hill resident who began research with Dr. Ming Lin last spring. His technical interests include digital audio, computer graphics, and math. As a musician for over 15 years, James has found digital audio to be a natural way to combine many of his passions.
Modeling Human Referring Expression Generation
Sahar Kazemzadeh, supervised by Tamara Berg
Referring generation expression is a natural language processing task that involves creating noun phrases that identify a referent object to a listener. We evaluate the state-of-the-art Visible Objects Algorithm for referring expression generation presented by Mitchell et al. (2012), and find that it does not perform as well with our natural image set as with the computer-generated image set that was originally used. Further, we analyze over 7,000 referring expressions generated by players of ReferIt Game, an online game that we developed, and by Amazon Mechanical Turk workers to identify metrics with which to create an improved stochastic model that can be coupled with computer vision to mimic human referring expression generation from visual input.
Sahar Kazemzadeh is a senior from Apex, North Carolina pursuing a Bachelor of Science in computer science with a minor in business. She is a recipient of the Herbert W. Jackson and North Carolina School of Science and Mathematics scholarships and a member of the Honors Program. During her four years here, she was a finalist for UNC’s premier entrepreneurship competition – the Carolina Challenge, received Governor McCrory’s Emerging Issues Award for Innovation, and was heavily involved with UNC Women in Computer Science as an officer. She is particularly interested in computer vision and natural language processing, namely for the challenging and exciting opportunities the fields present for merging the physical and digital realms. After graduation, Sahar will be joining Google as a software engineer.
Groups from E-mail Messages
Andrew Ghobrial, supervised by Prasun Dewan
Wouldn’t it be useful to have your e-mail client automatically generate relevant groups based on your past e-mail exchanges? This study aims to solve this problem by applying group generation techniques used in different contexts to e-mail. This study looked at data collected over twenty students and tested various techniques to determine which method generates groups that are most likely to be used in the future. The study applies Kelli Bacon’s group generation algorithm by generating different variations of graphs that represent the user’s past e-mail history. We conclude that Jacob Bartel’s bursty model and Google’s Interactions Rank formula produce groups that are most relevant to the user.
Andrew Ghobrial is a junior from Fayetteville, North Carolina majoring in computer science with minors in Arabic and entrepreneurship. He is interested in applying original computer science research to create viable technology start-ups. This fall, he will be studying at University College London.
Distributions-Based Approach to Predicting Answer Times on Stack Overflow
Preeti Arunapuram, supervised by Prasun Dewan
There are many online forums to which people go to have questions answered, such as Yahoo Answers, Stack Overflow, and Newsgroups. Oftentimes, the sender may urgently require a response to an important question. Being able to predict responses to online posts would help a person gauge whether or not he or she wants to wait for an answer to a question before moving on to another information source.
Preeti Arunapuram is a senior undergraduate from Philadelphia studying mathematical decision sciences and computer science. She is interested in machine learning, and she plans to attend graduate school in California next year.
Securing Data on Compromised Hardware
Matt Corallo, supervised by Michael Reiter
Advancements in attacks with physical access to commodity hardware has resulted in a general consensus that, given physical access, all of the data on a machine should be considered compromised. In this talk, I will consider existing attacks and mitigations and propose a practical system under which certain data security guarantees can be made. Specifically, I propose a system which attempts to protect encryption keys absolutely against physical attackers and uses them to protect both main memory and disk access.
Matt Corallo‘s research interests in computer science include security and distributed systems. A Bitcoin developer, Matt is graduating early to pursue research in advancing the flexibility and security of Bitcoin.