Undergraduate Research Poster Day 2012

April 16, 2012

<div style=\"margin-top:15px; margin-bottom: 20px;\">

<script type=\"text/javascript\" src=\"http://www.cs.duke.edu/web-resources/slideshow/m/embed.js\">

<div id=\"album-20\">

<script type=\"text/javascript\">
attributes: {
id: \"album-20\",
width: 515,
height: 373
mobile: {
auto: false,
poster: \"vignette\"
params: {
bgcolor: \"#ffffff\",
allowfullscreen: true
flashvars: {
xmlFilePath: \"http://www.cs.duke.edu/web-resources/slideshow/images.php?album=20\",
navAppearance: \"Numbers\",
transitionStyle: \"Cross Fade\",
feedbackPreloaderAppearance: \"Hidden\",
contentOrder: \"Random\"

<p style=\"margin-top:15px;\"><span style=\"font-size: 34px; float:left; line-height:100%; margin-right:3px;\">Ten computer science undergraduates presented research projects April 13 during the second annual Computer Science Undergraduate Research Poster Day.

Daphne Ezer, a Marshall scholar and CS undergraduate research fellow, was selected to receive �Best Poster� recognition for her project: �Mechanistic Model of How AT Tandem Repeats Influence Gene Expression and Evolution.�

�This year�s event was a great success,� Computer Science Associate Chairman Richard Lucic said. �These were excellent posters, and many faculty and graduate students stopped by to interact with the presenters.�

The Poster Day, held in the Levine Science Research Center�s Hall of Science, recognizes the achievements of participants in the C-SURF program and other Graduation with Distinction candidates. Undergraduates who successfully complete a research independent study with faculty and oral defense are qualified to graduate with distinction.

In addition to recognizing student achievements, it is hoped that the event also generates interest and enthusiasm among faculty and students for undergraduate research, Lucic said.

For more than an hour, the students named below stood by their posters describing their work as other students and faculty mingled through the area asking questions.

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_dorminey_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Sterling Dorminey

Sterling Dorminey, who worked with Computer Science Chairman Carlo Tomasi, worked on tracking people by using 3-dimensional sensors. The idea, he said, is akin to Harry Potter�s marauders map, a magical document that shows the location of individual people throughout the fictional Hogwarts campus. Dorminey�s project used sensors in combination with a computer program to track people as they move from sensor to sensor. He focused specifically on trying to figure out where people are in a given image from the sensor system.

�All the computer sees are a matrix of different pixels,� he said. �If you have two people that are right next to each other or one person�s behind another and you can�t see them all, then computers really freak out on that sort of thing.�

To deal with the problem, Dorminey had images separated based on a section�s level of darkness to come up with boundaries for objects and people. A support vector machine then was used to determine whether the boundaries in a given image were of people. Such data then would be passed on for correlation clustering, a project by fellow senior Trevor Terris, to ultimately track people for security purposes.

After graduation, Dorminey will work on a cloud platform for Microsoft, beginning in September.

View Sterling Dorminey\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_ezer_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Daphne Ezer

Daphne Ezer, a C-SURF Fellow who worked with Professor Alexander Hartemink, worked on a project in computational biology to try to predict whether genes from a DNA sequence are turned on or off.

�In front of some genes, there are repeating DNA sequences,� Ezer said. �I�m trying to predict how the length of these repeats could influence how proteins bind to the DNA and how that would influence how often genes get turned on or off. I�m also looking to see whether I can predict how the genes will evolve under stress.�

How to determine whether the genes would be turned on at high levels, based on the DNA sequence and the different proteins that potentially can bind to the DNA, is one of the big unsolved challenges in computational biology. For her project, Ezer looked particularly at genes in yeast that have small repeats, a group of genes that frequently include stress response genes. Short DNA repeats also are found in front of genes in humans. Ezer created a model to determine whether the length of the repeats influence how proteins bind to the DNA.

Ezer is validating her research with the hope of having her project published. After graduation, she will pursue a doctorate in the computational side of genetics at Cambridge University.

View Daphne Ezer\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_genkins_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Julian Genkins

Julian Genkins, a C-SURF Fellow who worked with Professor Susan Rodger, conducted research to improve the flexibility of JFLAP (Java Formal Languages and Automata Package), a program of graphical tools that help in learning basic concepts of formal languages and automata theory.

�My goal was to improve the flexibility of the program by refactoring the code and implementing new features, which allowed for greater flexibility in language,� Genkins said. �Specifically, this included allowing strings as symbols and providing explicit formal definitions.�

Preliminary testing indicates students enrolled in Duke�s FLA theory class, CS140, find the changes useful and the software more effective, he said.

After graduation, Genkins will work as a software developer for Epic Systems in Madison, WI.

View Julian Genkins\' poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_goh_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Ethan Yong-Hui Goh

Ethan Yong-Hui Goh, working with Professor Pankaj Agarwal, attempted to produce a scalable algorithm to simplify contours of a terrain surface.

�The advent of modern LiDAR technology has allowed for geographic data to be gathered en masse,� Goh said. �However, high-resolution measurements can cause programmatically generated contour maps to be large and noisy.�

Instead of trying to smooth existing contours to make them prettier, Goh explored simplifying the terrain�s underlying triangulation through edge operations to generate �nice� contours from the start.

�Generating contours directly from large datasets is not ideal,� he said, �because the final output might be too large for most applications to handle. Being able to reduce the output size so that contour maps can be used in other post-processing programs is crucial to the geographical information system.�

After graduation, Goh will work as a software developer for Microsoft.

View Ethan Goh\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_levy_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Joe Levy

Joe Levy, a C-SURF Fellow who worked with Associate Professor Romit Roy Choudhury, created a projected user interface with a mobile phone. The interface makes use of phones recently on the market with built-in projectors. By combining a phone�s projector and camera, content on the screen can be projected on a wall and manipulated in the same ways, allowing phones to be used as SMART Board interactive whiteboards or to give dynamic presentations. With a laser pointer, a small amount of information can be circled and seen full screen or text can be underlined and highlighted in a different color.

�I�ve always been interested in mobile phones,� Levy said. �I thought there were definitely some limitations to what you can do on a tiny screen and it�d be nice to improve the usability of cell phones. It�s not as easy to see things, it�s not as easy to manipulate things because we have big fingers and the screens are small.�

After graduation, Levy will work in the Servers and Tools Division at Microsoft.

View Joe Levy\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_narayan_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Trevor Narayan

Trevor Narayan, a C-SURF Fellow who worked with Associate Professor Romit Roy Choudhury, conducted research into solving the problem of the proliferation of mobile devices all transmitting high-quality images and video. Working under the assumption that humans care more for certain aspects of a photo, such as a baby pictured on his poster, Narayan�s group proposes that bandwidth can be conserved by extracting and transmitting the focus of the photo and then substituting the background with a template already on the recipient�s mobile device.

�The point is not to trick the mobile user into thinking that they�re getting the original image,� Narayan said. �You�re more just looking for acceptability. Is this acceptable given bandwidth constraints? We�re sort of looking for a compromise. This is just one technique to try and solve what�s going to be a global problem at some point.�

The solution has not been implemented successfully yet, however, as his group has found it is difficult to clearly identify what people will think is important in a photo.

After graduation, Narayan will work as a software consultant for Red Hat in Mountain View, CA.

View Trevor Narayan\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_schmidt_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Tanner Schmidt

Tanner Schmidt, a C-SURF Fellow who worked with associate professors Jeffrey Forbes and Ronald Parr, researched 4-dimensional mapping with mobile robots and RGB-D sensors. The goal was to use a robot and depth cameras to map an environment over time, creating a series of maps to determine changes in the environment, such as a door that suddenly opens when it�s usually closed from midnight to 8 a.m.

�You can use it to improve current navigation techniques,� Schmidt said, noting robots could use stored histories of an environment to plan routes or possible detours. �And there can also be security applications. Based on the history of a map you can determine when you observed a change and how probable is this change.�

A closed door could be a high probability change as doors open and close all the time, he explained, but a computer removed from a desk that normally never moves might signal the need for a security check.

After graduation, Schmidt will pursue a doctorate in computer science at the University of Washington, likely focusing on artificial intelligence.

View Tanner Schmidt\'s poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_scripa_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Andrea Scripa

Andrea Scripa, a C-SURF Fellow who worked with Associate Professor Jeffrey Forbes, worked on mining short texts on Twitter for information. Specifically, she researched how to analyze information received from tweets for situational awareness. Tweets from people involved in a large-scale event, such as a natural disaster, could be useful for providing immediate information to emergency response teams even before news media begin reporting on an emergency. Yet, Scripa said, �the current solutions we have for processing unstructured documents are really for much longer texts that don�t contain irregularities like slang or misspellings or strange emoticons. The bodies of text found on microblogs resemble common speech patterns much more closely than they do formal prose, and this makes for a much different problem.�

After determining that six current spatial and statistical methods for classifying information work similarly, she looked at how to bring in other information � such as user, social network or linguistic information � to enrich the data and classify it better. Her results were mixed. �We haven�t found a model that we can objectively say (by scientific measures) is the best, and some forms of data enrichment work better than others,� she said. �Basically, your optimal model selection and the ideal subset of metadata to incorporate are highly dependent on the specifics of the problem you\'re trying to solve. Since currently there is no clear, overall winner for classifying situational awareness, there is definitely room for more improvement in the future.�

After graduation, Scripa will work as a program manager for Microsoft in Research Triangle Park.

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_terris_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Trevor Terris

Trevor Terris, a C-SURF Fellow who worked with Computer Science Chairman Carlo Tomasi, worked on identity management using correlation clustering. The goal was to be able to identify from given video footage sets of observations of individual people. To try solving the problem, he mapped images to a graph, using nodes to represent every observation of a person and weight edges to represent the level of confidence that individual images are of the same person. Two people in the same frame were assigned an edge weight of negative infinity, showing 100 percent confidence they are not the same person, while two similar-looking people in images close together in space and time were given a positive edge. Terris also created an algorithm to quickly cluster images into sets most likely to be of the same person. More work, however, is needed to solve the problems of finding people in each frame and assigning edge weights. Once those problems are solved, Terris said, the correlation clustering could be used as a partial solution for tracking people in airports or subway systems.

�The applications are huge,� he said. �It just will take a lot more research before it gets there.�

After graduation, Terris will consult with Red Hat in Raleigh, beginning in August.

View Trevor Terris\' poster

<div class=\"OneNewsPhoto\" style=\"width:95px; float:left; margin-top:5px; margin-right: 20px; margin-bottom:10px;\">
<img src=\"/web_resources/photos_news/398_yan_150x200.jpg\" class=\"NewsPhoto\" style=\"width:95px;\"/>
<p class=\"NewsPhotoCaption\">Jiaqi Yan

Jiaqi Yan, a C-SURF Fellow who worked with Associate Professor Jun Yang, conducted research into implementing statistical analysis on a massive scale inside a database system.

�Our purpose is to refine and enhance the existing database system,� he said, noting that just in the Levine Science Research Center, various labs work with large amounts of data all the time. Yet the computational power available today cannot efficiently handle all the data.

His approach to the problem is �chunk-oriented storage and execution,� a platform that allows support for different operations by assembling data in various shapes and sizes rather than the traditional focus on one set of operations at a time. He then uses a graphics processing unit to accelerate analysis by conducting parallel executions. The result is the computational time no longer is the bottleneck in analyzing data.

�It�s the assemble time that becomes dominant in the processing,� Yan said.

After graduation, Yan plans to pursue a doctorate in computer science at the University of Michigan.