Meeting of the Minds 2010

Page 1

Undergraduate Research Symposium

2010



About

Undergraduate Research Symposium Meeting of the Minds is an annual symposium at Carnegie Mellon University that gives students an opportunity to present their research and project work to a wide audience of faculty, fellow students, family members, industry representatives and the larger community. Students use posters, videos and other visual aids to present their work in a manner that can be easily understood by both experts and non experts. Through this experience, students learn how to bridge the gap between conducting research and presenting it to a wider audience. A review committee consisting of industry experts and faculty members from other universities review the presentations and choose the best projects and posters. Awards and certificates are presented to the winners.


ABOUT CARNEGIE MELLON UNIVERSITY With more than a century of academic excellence and distinguished research, Carnegie Mellon University is a leader in education with real-world applications. Consistently top ranked, the University has more than 11,000 students, 84,000 alumni and 4,000 faculty and staff globally. Core values of innovation, creativity, collaboration and problem solving provide the foundation for everything we do. At the invitation of Qatar Foundation, Carnegie Mellon joined Education City in 2004. Here, Qatar Foundation created a unique center for scholarship and research that is the ideal complement to Carnegie Mellon’s mission and vision. Students from Qatar and 30 different countries enroll at our world-class facilities in Education City. Carnegie Mellon Qatar offers undergraduate programs in business administration, computer science and information systems. Learn more at www.qatar.cmu.edu.


Table of Contents POSTER Q1 Q2 Q3 Q4

TITLE Business Administration posters An Approach to Improve the Future of Entertainment Family Business in Qatar: Challenges and Sustainability Futuristic Hospitality Hot Seat Solution

PAGE 2 4 6 8

Q5 Q6 Q7 Q8 Q9 Q10 Q11 Q12 Q13 Q14 Q15 Q16 Q17 Q18 Q19 Q20 Q21

Computer Science posters Arabic Lip-Syncing for 3D Face Animation Applications Autonomous Race Car Control Optimized Using a Genetic Algorithm Becker’s ++: A Fun and Effective Way to Learn Programming Compute the Distance Between K-Dimensional Vectors on Wimpy Nodes Dynamic Path Planning and Traffic Light Coordination for Emergency Vehicles Education E-Village: Empowering Technology Educators in Developing Regions Effective Cost Estimation for Market-Based Multi-Robot Coordination Human Annotations Evaluation I-GEST A Multi-Point, Low-Cost, Gesture-Based Input Platform Mobile Phone-Based Educational Games to Improve Adult English Literacy Recognizing Places Using Image Recognition Rich Entity Type Recognition in Text Statistical Language Translation into Morphologically Complex Languages The Quantum Computer Alternative of a Common Algorithm Training a Roboceptionist to Predict Input Classes Using Eigenfaces to Guess Social Context Your voice as your ID: A Biometric Approach

12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44

Q22 Q23 Q24 Q25 Q26 Q27

Information Systems posters Carnegie Bazaar Curriculum Development and Reflections on Teaching English to Migrant Workers Fisheye KMS: A Manager’s View of a Technology Roadmap ‘Onigiri’ or How IT Can Raise Awareness on Diabetes StudentTRAK Transactional Referral System

48 50 52 54 56 58

Q28 Q29 Q30 Q31 Q32 Q33 Q34 Q35 Q36 Q37 Q38 Q39

Post-graduate posters Assistive Educational Technology Project Coverage Problem in Wireless Multimedia Sensor Networks Hala the Bilingual Robot Receptionist Initial Findings on Provisioning Variation in Cloud Computing Interference in Multihop Wireless Networks LNG Pipe Vision Open Domain Named Entity Recognition for Arabic Student Modeling for the Adaptive Braille Writing Tutor Task Allocation and Coordination in a Disaster Response Scenario Task Allocation and Scheduling for Constrained Teams Touch-Screen Interface Design for Multicultural Settings Wireless Network Optimization Using Software-Defined Radios

62 64 66 68 70 72 74 76 78 80 82 84



Business Administration posters


An Approach to Improve the Future of Entertainment

Poster Q1

Authors Maryam Al-Kuwari (BA 2011) Hala Khashabi (BA 2011) Faculty advisor Sham Kekre, Ph.D. Category Business Administration Abstract Overall, there are limited choices for entertainment in Qatar, and even the facilities available have opportunities for improvements. In this research, we are focusing on the movie theater industry as we found that going to the movies is one of the main sources of entertainment in Qatar. After conducting a survey, we found out that only four percent of the sample attend movies at the Mall Cineplex. As a result of that very small percentage, we decided to focus on this cinema as we see opportunities to make it a successful enterprise in the future. We are focusing on why people prefer other cinemas such as Villagio and City Center, and how can we improve the service offerings in the Mall Cineplex in order for it to compete effectively with other competitors in the market. The poster will include the problem, the reason behind choosing this particular cinema, the solution and recommendations.

2



Family Business in Qatar: Challenges and Sustainability

Poster Q2

Author Saad Al-Matwi (BA 2010) Faculty advisor Sham Kekre, Ph.D. Category Business Administration Abstract My research project is focused on family business in Qatar. Family businesses are playing a major role currently in developing Qatar’s non-oil economy, primarily in the service sector. They also were instrumental in the initial oil and gas sector development. Therefore, a study of family business will help us to understand how the State of Qatar came to be and foresees the kind of the challenges it might face in the future. I conducted a field survey with owners, managers and experts in family business in Qatar to determine the main elements of their business and the key challenges they face. This research is important for many reasons. First, there is a lack of a field surveys on Qatari family business (unlike other family business around the world). Second, Qatar is entering new fields of development in its attempts to diversify its economy with a focus on areas such as education, health, tourism and finance. Thus, it is important to see how these businesses will react to these new changes. These are also areas where family businesses provide support services. Third, many family business in Qatar are in the second or third generation phase. This phase is crucial for family business, as we know that in the rest of the world many don’t make it to the fourth generation. Finally, my goal in this research is to survey the family business world in Qatar, identify its challenges and propose recommendations for its sustainability. The study of family business is an emerging field in the world of business education. Many of the methods that are applied to corporate company can’t be applied to family business. In addition, family businesses develop unique approaches in dealing with business challenges that can be leveraged in corporate companies. Furthermore, the study of family business is essential for anyone individual who wants to understand the GCC economies.

4



Futuristic Hospitality

Poster Q3

Authors Benazir Anis (BA 2011) Mehrunissa Anis (BA 2010) Arsalan Arif (BA 2010) Faculty advisor Sham Kekre, Ph.D. Category Business Administration Abstract Our poster explores the dynamics behind a hotel service and how the pieces fit together, both behind the scenes and in front. We want to explore what it takes to make a good experience, and thus a satisfied customer, in a hotel stay. Research topics include finding out about successful hotels in the world. We looked at a range of prices and star levels to eliminate the factors of price and focus on the service. We pursued our research and look at local or regional hotels and how they differ from the more innovative and successful hotels and see if those can be incorporated here, and how they can be improved for the future. We will show some examples of innovative hotels and how you can use issues such as sustainability, cultural differences and technology to create a great hotel for the future, and not depend on international brands. We came up with the idea as part of our course “Managing the Enterprise of the Future� taught by Professor Sham Kekre. We looked at issues regarding the traditional service methods and how firms employing new and innovative service methods become successful. We primarily used cases and the Internet as well as our own experiences of hotels around the world. There are some great Web sites that cater to innovative and sustainably designed hotels of the future. This is extremely important because the growing economies of the region are increasingly diversifying their operations from oil & gas and focusing on the service sector. An important part of the sector is tourism and more tourism means more hotels. For a small country like Qatar, it will be very beneficial to differentiate with services and focus on sustainability.

6



Hot Seat Solution

Poster Q4

Author Laila Murad (BA 2011) Faculty advisor George M. White, Ph.D. Category Business Administration Abstract Hot Desking is a very attractive work environment where no employee has a certain work space rather the whole space is shared between all different levels of employees. Hot Desking has proved to be a very effective working environment that motivates the employee to perform at his/her highest level. However, it is not very commonly adopted because of its limitation. Limitations such as inefficient use of work time and lack of effective work related communication among the employees. My alternative, “Hot Seat Solution,� solves such limitations by making communication between the employees far more effective and efficient. And it creates a great advantage in using the time that was previously wasted more effectively. Adopting this solution gives companies and corporations a real chance to experience a very unique and interactive work environment that combines the best of both worlds. So it is the fun environment, which allows employees to engage with one another creating a work family away from home, and it a work place in which work is done professionally and in a timely manner.

8



10


Computer Science posters


Arabic Lip-Syncing for 3D Face Animation Applications

Poster Q5

Author Ossama Obeid (CS 2012) Faculty advisor Majd Sakr, Ph.D. Category Computer Science Abstract Three dimensional face animations have been widely used mostly in gaming and entertainment, and more recently are finding their way into practical applications. A problem that we face today is the lack of a specialized lip-syncing model for the Arabic language. A good model would describe the smallest number of different lip shapes needed to achieve realistic (ideally) Arabic lip-syncing. Further, mapping existing lip shape models from other languages, limits the realism that Arabic animated faces can achieve, and thus decreases the effectiveness of the underlying application using the animated face. Current approaches consist of borrowing an animation model from other languages (mostly English) and finding the closest approximation of an Arabic mouth shape/transition. This turns out to be ineffective since the English mouth shapes do not encompass all the Arabic mouth shapes. Furthermore, few languages match the morphemes of the Arabic language, which makes borrowing from other languages ineffective. Our goal is to build an Arabic lip-syncing model via experimental methods. This model consists of a minimal list of mouth shapes needed to describe the Arabic morphemes. Through analyzing the Arabic phonemes, we conjecture that a reduced list is possible. To evaluate our conjecture, we capture videos of multiple volunteers uttering all the Arabic phonemes. We then analyze these videos, searching for similarities between the utterances of each phoneme among all the volunteers to develop a minimal list of mouth shapes as our model. Then we experimentally test the effectiveness of our observed model. These tests consist of getting other volunteers to look at some of the previously recorded footages (muted) and try to identify what phoneme is being said. We check whether the volunteers’ answers are consistent with our model.

12



Autonomous Race Car Control Optimized Using a Genetic Algorithm

Poster Q6

Author Mohammed Janahi (CS 2012) Faculty advisors M. Bernardine Dias, Ph.D. Brett Browning, Ph.D. Category Computer Science Abstract Complex optimization problems that include large numbers of parameters are often best solved with local search techniques such as Genetic Algorithms. Many of these problems, such as optimized control of complex mechanisms, or optimized scheduling of operations in large production plants, are both important and ubiquitous in society. Genetic Algorithms solve these problems by evolving populations of solutions through operations such as reproduction, crossover, and mutation. While Genetic Algorithms present a powerful mechanism for solving complex problems, their major drawbacks are that they cannot guarantee an optimal solution, and that they are not easily customized to solve a specific problem effectively. This project explores the effective use of a Genetic Algorithm to optimize 50 control parameters to maneuver a racecar on its course. We implement and customize a Genetic Algorithm for this problem, and use The Open Racing Car Simulator (TORCS) to evaluate the solutions. We also study the change in solution quality when the Genetic Algorithm parameters such as generation size and mutation probability are altered. Overall, we show that a Genetic Algorithm can optimize control parameters to effectively maneuver a racecar on its course.

14



Becker’s ++: A Fun and Effective Way to Learn Programming

Poster Q7

Authors Samreen Anjum (CS 2011) Rishav Bhowmick (CS 2010) Faculty advisors Hoda Fahmy, Ph.D. Saquib Razak, Ph.D. Category Computer Science Abstract There have been many tools developed to aid instructors in teaching programming languages. Each tends to follow different approaches to keep students active in their learning process. One tool, Becker’s Robots, is aimed at teaching programming in a creative and engaging way. It enables students to learn procedural logic within the context of an object-oriented paradigm. The purpose of this work is to extend the Becker library and provide prospective instructors with a database of demos and exercises to ease the introduction of the fundamental programming concepts. The current Becker library supports exercises with a robot theme. The main goal of our project is to extend the library so that the following themes are supported:

• SoccerBot: A soccer field with soccer robot and a soccer ball that simulates the soccer game and rules

• CatchMeIfYouCan: A maze city where a thief’s main objective it to steal the diamonds and hide from the police, while the main goal of the police is to chase the thief.

• ShopOut: A store where a doll can try on clothes and purchase merchandise.

After interviewing a number of high school teachers as well as freshman instructors, we have started to work on three motifs that target the diverse student population here in Qatar. Preliminary results are encouraging and show that students are more engaged. We are working on developing ways to further evaluate the effectiveness of our initiative.

16



Compute the Distance Between K-Dimensional Vectors on Wimpy Nodes

Poster Q8

Author Yi Luen (Tessa) Eng (CS 2010) Faculty advisor David Andersen, Ph.D. Category Computer Science Abstract Computing distances between high-dimensional feature sets is an important problem, with applications in computer vision, machine learning, computer security and more. It is now being used as the basis for cluster-based applications, such as the scene completion problem below. In this context, power and computational efficiency are gaining importance: recent EPA estimates suggest that 50 percent of the total cost of owning a computer will be due to power and cooling. Low-power systems are gaining popularity for use in clusters and I am trying to investigate how well an important workload can run on these systems. In this research, we used various techniques of optimization on James Hays’ Ph.D. thesis, “Scene Completion Using Millions of Photographs,” which required computation, including L1 & L2 distance and chisquare, is both I/O and CPU intensive: It performs multiple vector comparisons against every image in a huge set downloaded from Flickr. The goal of this project is to optimize the speed of the computation of distance between vectors and analyze the cost of running the computation on wimpy nodes. We present our optimization approach and corresponding results for the computation of distance between vectors.

18



Dynamic Path Planning and Traffic Light Coordination for Emergency Vehicles

Poster Q9

Author Yi Luen Tessa Eng (CS 2010) Hend Gedawy (CS 2009) Faculty advisors M. Bernardine Dias, Ph.D. Khaled Harras, Ph.D. Category Computer Science Abstract The result of an emergency vehicle, such as an ambulance or fire truck, arriving a few seconds late at an emergency site can be the difference between life and death. As different technologies emerge, various approaches to minimizing emergency vehicle travel time have also evolved. Our research builds on the state-of-the-art in vehicle path planning to help further reduce this travel time for emergency vehicles. In our work, we combine dynamic path planning with traffic signal pre-emption to achieve this goal. D*Lite, an informed search algorithm, efficiently and optimally plans and replans according to changing costs and travel times of routes in the traffic network. After a route is chosen, a pre-emption strategy is applied to reduce emergency vehicle delay while maximizing traffic flow through the intersection. We evaluate our solution through analytical experiments using our implementation of D* Lite, and through scenarios developed using the VISSIM specialized microscopic traffic simulator. The results validate our hypothesis demonstrating that dynamic path planning can improve travel time under uncertain congestion conditions, and that incorporating an appropriate traffic light pre-emption mechanism can further improve travel time for an emergency vehicle; potentially saving lives. We present our approach and simulation results in this poster.

20



Education E-Village: Empowering Technology Educators in Developing Regions

Poster Q10

Author Kaleem Rahman (CS 2010) Faculty advisor M. Bernardine Dias, Ph.D. Category Computer Science Abstract There exists a significant need for relevant, accessible and useful resources to enhance technology education in developing regions. Currently, access to courseware for technical subjects such as Computer Science is available via several online resources. However, these resources are designed for developed communities where technology is ubiquitous, technology infrastructure is robust and educators have easy access to a variety of academic publications and other helpful guides. Therefore, the available online resources do not provide sufficient avenues for educators in developing regions to understand the courseware or discuss alternative ways of teaching it based on their specific constraints. To address this deficit, the TechBridgeWorld group at Carnegie Mellon University initiated the “Education e-Village� (E-Village) project. E-Village is an online community where educators from around the world will be able to share ideas, experiences, expertise, educational resources and strategies to promote and enhance technology education in developing regions. This senior thesis project enhances the search functionality and user experience of the E-Village initiative. The target users of E-Village span educators in both developed and developing regions who vary widely in terms of their exposure to technology and access to resources. Thus, the search functionality and user experience of E-Village must cater to this diverse set of expectations and constraints in order to successfully engage and retain the community of users. In this work we conducted a set of user studies with potential future E-Village users to understand their relevant needs and constraints. We accordingly selected and configured the Lucene search engine for E-Village, evaluated and improved the existing user interface to E-Village, and enhanced the overall user experience in E-Village.

22



Effective Cost Estimation for Market-Based Multi-Robot Coordination

Poster Q11

Author Ahmed Emam (CS 2012) Faculty advisors M. Bernardine Dias, Ph.D. Brett Browning, Ph.D. Category Computer Science Abstract Effectively coordinating a team of robots to execute a set of tasks in dynamic and uncertain environments is a challenging problem. Yet, this problem must be solved efficiently for robots to play a useful role in society. Many future applications, including automated construction, hazardous exploration and disaster response will require teams of robots to work together efficiently despite many uncertainties and imperfect conditions. While several approaches have been proposed to address this coordination problem, market-based techniques have become an increasingly popular solution over the past decade due to their robustness and efficiency under dynamic conditions. Marketbased approaches model the coordination problem as a virtual economy where tasks are allocated using auctions in which the robots are the traders and the tasks are the commodities being traded. Robots estimate the costs of executing tasks according to an objective function such as time or distance. Accurate cost estimation for bid valuation and task scheduling therefore is a central component of a market-based solution. This project explores using the A* algorithm to estimate distance-based costs for a team of robots performing a disaster response task. These cost estimates are used for bid formulation and task scheduling in the TraderBots market-based coordination approach. Team performance is measured as the sum of the distance traveled by each robot on the team. We evaluate our A* approach by comparing performance with cost estimation based on Euclidean distance. Our results demonstrate that A* is an effective cost estimation technique for market-based coordination of a team of robots.

24



Evaluation of Human Annotation

Poster Q12

Authors Reham Al Tamime (IS 2013) Mariem Fekih (IS 2013) Faculty advisor Behrang Mohit Category Computer Science Abstract Computer processing of languages is a challenging task due to the diversity of the languages and the complexities of their structures. We work on a project that trains a machine learning system to recognize and classify named entities in Arabic text. The machine learning system requires labeled text with named entities annotated. Our first task was to annotate different articles from Arabic Wikipedia. For each article we first introduced two to three domain-specific classes. Moreover, we tagged the text with the named entity information. Tagging included locating and categorizing basic named entities - person, organization and location - and also the article-specific classes. In order to increase the performance of the system, there should be a consistency of annotation. We measure the annotation consistency by an agreement metric. To do that, we wrote a Java program that calculates the agreement percentage of two annotators. Using our program, we conducted several tests and analyzed different aspects of disagreement in two stages of annotation. As an extended work, we are implementing other agreement metrics (e.g., Cohen’s Kappa) that take issues such as the randomness of tagging into account. This work is part of the American and Qatari Modeling of Arabic (AQMAR) project that is funded by QNRF through the NPRP fund (08-485 -1-083).

26



I-GEST A Multi-Point, Low-Cost, Gesture-Based Input Platform

Poster Q13

Authors Jackson Davis (CS 2010) Swapnil Joshi (CS 2012) Faculty advisors Khaled Harras, Ph.D. Majd F. Sakr, Ph.D. Category Computer Science Abstract In this work we present I-GEST, a platform for the research, development and evaluation of low-cost, gesture-based intuitive computer input systems. The platform is coupled with applications evaluating usability that can potentially impact user interface design. The overall system includes infrared transmitting devices, infrared sensors, gesture recognition, software functionality and a set of simple applications. A user generates a gesture from an infrared pen as a transmitting device that is then received by a Wii-mote, an affordable infrared receiving device. This input data is then parsed by the I-GEST platform, which then interprets the corresponding gesture. This gesture can then be tied into various software applications that would ultimately be taking advantage of the user’s natural input. The software platform is easily extensible to enable future research and easy integration in software applications. This simple gesture-driven input platform can also be utilized to assess and evaluate usability for image manipulation, data-entry and basic communication applications.

28



Mobile Phone-Based Educational Games to Improve Adult English Literacy

Poster Q14

Author Aysha Siddique (CS 2010) Faculty advisor M. Bernardine Dias, Ph.D. Category Computer Science Abstract English is one of the most commonly used languages in international business, and therefore, some level of fluency in English becomes a pre-requisite for many employment opportunities. Due to their limited English proficiency and a lack of opportunity to improve their English skills, a variety of adult populations are disadvantaged in many ways, including career advancement and societal acceptance. For example, low-skilled immigrant workers in countries such as Qatar and the USA have limited English proficiency, which is often a barrier to their career advancement and creates communication problems with their supervisors. Similarly, limited English skills make it harder for refugee populations to find jobs and adjust to the local culture in their host countries. Also, the average deaf adult in the USA reaches only a 4th grade English reading level. Our work aims to address the problems of limited English proficiency among adults by providing these groups with a low-cost, easily accessible, fun tool for enhancing their English skills. Mobile phones are the most prevalent and accessible computing technology for people of all ages and incomes. Related research efforts by several groups have demonstrated the success of mobile phone-based educational games in improving English literacy skills among primary school students. The goal of our work is to investigate the effectiveness of mobile phone-based educational games on adult English literacy. Our literacy tool consists of two parts: a single player game accessible on a mobile phone, and an online content authoring system that enables teachers to add useful educational content to the games. We incorporate proven techniques from expert teachers into these educational games, along with graphics and game concepts that motivate adults to play these games. The combined result is an effective and ubiquitous tool for enhancing English literacy skills among adults with limited English proficiency.

30



Recognizing Places Using Image Recognition

Poster Q15

Author Ossama Obeid (CS 2012) Faculty advisors Brett Browning, Ph.D. Bernardine Dias, Ph.D. Category Computer Science Abstract Reliably recognizing places is a key navigation task for robots and people alike. For robots, recognizing a place is essential to robust mapping and navigation particularly in GPS denied environments. In this project, we explore two techniques for place recognition from imagery. The first method examines simple low-resolution color images to represent places while the second uses a Bag-of-words model built over local features extracted using the SIFT algorithm that are then quantized using K-means. We evaluate the performance of the algorithm for place recognition using video data extracted from a vehicle driving around the Education City area. Images were pose tagged with GPS locations enabling quantified comparisons of place recognition performance.

32



Rich Entity Type Recognition in Text

Poster Q16

Author Rishav Bhowmick (CS 2010) Faculty advisors Michael Heilman Kemal Oflazer, Ph.D. Noah A. Smith, Ph.D. Category Computer Science Abstract Software applications built for summarizing a document or answering questions would greatly benefit with an Entity Recognizer. Today, there is a need to indentify entity (nouns) in free text. Although it might seem trivial for humans to recognize the entities in a sentence, it is a challenge for a machine to identify them. Furthermore, it is not feasible to list all instances of single entity type. For example, it is not simple to enumerate all possible entities of type “Person.� Adding to the challenge is the ambiguity in the language. For example, Washington could refer to a city, state in the USA or even a name of a person such as George Washington. Also, complex noun phrases make it harder for the machine to recognize (E.g.: mountain bike). Finally, lack of annotated data poses a major problem in training a recognizer. The goal of this senior thesis is to improve the performance of an existing Entity Recognizer called the Supersense Tagger. Supersenses are broad semantic classes, from Wordnet, used to tag words. The way we approached the problem is by looking at which existing features of the tagger affect the performance. We also varied the training data size to see if more training data really matters. We added an additional feature of word cluster and increased the context size (how many words before and after the current word being tagged). In this work we will illustrate the performance of the extended supersense tagger in recognizing entities in English text.

34



Statistical Language Translation into Morphologically Complex Languages

Poster Q17

Authors Dania Abed-Rabbou (CS 2012) Ahmed Emam (CS 2012) Mohammed Janahi (CS 2012) Faculty advisor Kemal Oflazer, Ph.D. Category Computer Science Abstract Automatic language translation is a computationally demanding task and has been tackled with multiple methods and solutions. Recently statistical approached based on parallel translated task have been very popular. In this project, we are trying to apply new approaches for statistical machine translation to translate from English to a morphologically complex language, Arabic. The process involves enumerating morphological patterns in Arabic and identifying corresponding syntax patterns in English. We can then perform syntactic analysis on English sentences and morphological analysis on the Arabic sentences and transform English sentences so that they are close to each other in word order. We can then use factored phrase-based framework to build a statistical translation system.

36



The Quantum Computer Alternative of a Common Algorithm

Poster Q18

Author Ossama Obeid (CS 2012) Faculty advisor Yonina Cooper, Ph.D.. Category Computer Science Abstract Joint Density Estimators (JDEs) are an accurate way of describing the joint distributions of discrete variables, and are frequently used for small applications in Artificial Intelligence (AI). However, with current technology, the memory used to store the joint distributions increases exponentially as the number of variables increase, thus rendering their use inefficient. I present a JDE counterpart for quantum computers in which memory usage grows linearly with the number of variables. Assuming the joint distributions have been previously calculated, and the variables used take on binary values, and also considering some constraints of quantum computing, one can reduce the memory used by the JDE by constructing a quantum gate that takes an n-qubit register, where n is the number of variables in the JDE, and transpose that qubit system such that, when observed, the resulting values may occur with the same probability as the given JDE. I also demonstrate how this quantum JDE would eventually be used if it were be implemented on a quantum system. Quantum computers seem very far off from making a public appearance. However, the work presented here gives more insight on how quantum computers can change our perception of classical algorithms, in this case, the performance of JDEs. This work also shows how Artificial Intelligence can benefit from both the randomness and the probabilistic nature of quantum computers since both are very common in modern AI applications .

38



Training a Roboceptionist to Predict Input Classes

Poster Q19

Author Raggi al Hammouri (CS 2011) Faculty advisors Brett Browning, Ph.D. M. Bernardine Dias, Ph.D. Category Computer Science Abstract Human Robot Interaction (HRI) is a challenging field that has broad potential impact on how robots interact with people. Good human robot interaction requires deep natural language and contextual understanding capabilities. In this work, we investigate the use of supervised learning techniques to improve the ability of a robot receptionist to recognize the context of its conversations with humans. Hala is a roboceptionist at Carnegie Mellon University in Qatar. She is designed to answer questions and engage the people around her in English and Arabic. To understand the users input, Hala currently uses hand crafted decision rules that operate over the last input sentence. We explore the use of classifiers, learned from hand labeled logs from over a week of Hala’s interactions, to classify the users recent history of input into predefined categories. We evaluate the performance of the classifiers on unseen data as a way to evaluate the effectiveness of the techniques.

40



Using Eigenfaces to Guess Social Context

Poster Q20

Author Samreen Anjum (CS 2011) Faculty advisor Justin Carlson, Ph.D. Category Computer Science Abstract Human-robot interaction is a rapidly evolving field that seeks to make human communication with robots more intuitive and efficient. Gender markers are pervasive in human language. To communicate “naturally,� a robot needs to be aware of the gender of the person with whom it is communicating. The goal of this project is to use vision to enable a robot to categorize humans based on gender. Our approach leverages the well-known principal component analysis technique from Turk and Pentland called Eigenfaces. Eigenfaces can be used to summarize the significant features of a face, and have in the past been applied to the tasks of face detection and recognition. In this project, faces found by the robot are projected into a reduced-dimensionality subspace spanned by these Eigenfaces, known as a face space. A set of faces for which gender is known is used to train Support Vector Machine and AdaBoost classifiers. These classifiers learn which portions of face space correspond to women and which portions correspond to men, enabling the robot to generate gender classifications for faces not seen before in the system. In this poster, we present our procedure, our results using the different classifiers and our insights into which components are most informative for determining gender, particularly in an Arab society.

42



Your voice as your ID: A Biometric Approach

Poster Q21

Author Tarkay Jamaan (CS 2012) Faculty advisors Brett Browning, Ph.D. M. Bernardine Dias, Ph.D. Category Computer Science Abstract Voice identification is an active field of research in computer science and biometrics. One of the big motivations for research is the emergence of ubiquitous computing. In a future where humans command machines by their voices, a machine has to be able to identify who issued the command in addition to what was said. In this poster we try to solve this problem using a biometric approach. Our system records certain key phrases and uses that information to identify the speaker. Following other approaches in music theory, signal landmark detection, computer vision and speech recognition, we process the sound data using the Fast Fourier Transform (FFT) along with other signal processing techniques. To classify the sounds we collected and labeled a data corpus that we used to train a Support Vector Machine (SVM) classifier on part of the dataset. We report on the evaluation performance of the SVM classifier on the remainder of the unseen data.

44



46


Information Systems posters


Carnegie Bazaar

Poster Q22

Authors Samee Amin (IS 2011) Shashank Jariwala (IS 2011) Omar Shaat (IS 2011) Faculty advisor Ian Lacey, Ph.D. Category Information Systems Abstract The application we are developing, The Carnegie Bazaar, is a classic example of how several factors concerning people, process and technology interact and play an important role in any software development project. The idea sprung as a means of providing a platform for Carnegie Mellon Qatar students to sell and purchase their used course books. However, as we looked deeper into defining a vision statement for our application, we realized that the proposed system has a lot of potential to serve the students in more ways. Hence, we broadened the project scope to include, but limiting to, transactions of books, electronics, furniture and any other tangible items not just amongst Carnegie Mellon Qatar students but also students from all the universities under Education City. Furthermore, after performing numerous user tests as well as surveys, we are extremely positive about the viability and necessity of such an application.There is a high demand for the existence of such an application since the entire Education City has lot of valuable commodities to offer for sale but no means to do so. And we believe our application will provide them the means to do just that. However, the goal of the project is not just to develop an online “souq” of sorts system for students of Education City. The team members have a great deal to learn from the project about the various aspects of project management and teamwork. Having team members based at Carnegie Mellon’s main campus in the USA has added a whole new dimension to this “Global Project” and offered an opportunity to experience challenges encountered in real life Information System projects around the globe.

48



Curriculum Development and Reflections on Teaching English to Migrant Workers

Poster Q23

Authors Mohammed Al-Hajjaji (MIS-SPC) Alreem Al-Nabti (MIS-SPC) Ghada Al-Sooj (IS 2012) Yasser Al-Salamy (MIS-SPC) Corinne Walters (CS 2011) Faculty advisor Silvia Pessoa, Ph.D. Category Liberal Arts and Sciences Abstract This poster reports on student participation in and curriculum development for a community literacy program for low-income migrant workers in Qatar. The 16-week literacy program called RAEL (ROTA Adult English Literacy) emerged of a partnership between Reach out to Asia (ROTA) and a major local company in Qatar interested in providing services to its employees through its corporate responsibility initiatives. Education City students who enrolled in a course on community service learning taught the once-a-week English classes. The course prepared them for teaching and helped them reflect upon their experiences. Although a theme-based communicative curriculum for teaching English for low-income migrant workers in Qatar was carefully designed for the purposes of this program, throughout the program the participating students often redesigned the activities in the curriculum and designed new activities to better meet the learners’ backgrounds, needs, levels and interests. This presentation offers an introduction to the literacy program, describes successes and challenges, and showcases teaching materials designed by the participating students to better meet the goals of the program. The presenters also reflect upon their engagement and interaction with the low-income migrant workers in the program and how this experience impacted them.

50



Fisheye KMS: A Manager’s View of a Technology Roadmap

Poster Q24

Author Varun Arora (IS 2012) Faculty advisor Iliano Cervesato, Ph.D. Category Computer Science Abstract This research project is intended to develop an information management system that provides SMEs, or Subject Matter Experts, with a user-friendly web-based editing and maintenance environment for the Technology Roadmap of the organization. This system will also act as a decision support system for the management allowing them to retrieve information selectively and make decisions based on it. This Fisheye Knowledge Management System (KMS) will embed elements from modern wikis, such as allowing SMEs to collaborate to create and update a technology article and providing fine control on accessibility, publishing, etc. Such control will allow information sharing between key stakeholders across the organization. Differently from wikis, it will also allow management to view sections of interest across all or part of the articles. The KMS will allow SMEs to enter various kinds of textual and visual information in editing environments for individual sections (of the article), which describe the nature and content of the information. This structure of the editing environment will be standardized, yet extensible, which will result in a common layout across all the articles.

52



‘Onigiri’ or How IT Can Raise Awareness on Diabetes Poster Q25

Authors Nur Aysha Anggraini (IS 2013) Sumya Khandaker (IS 2013) Maryam A. Yousuf (IS 2013) Faculty advisor Selma Limam Mansar, Ph.D. Category Information Systems Abstract The prevalence of diabetes in Qatar is alarming; so much as it is around the world. According to the Qatar Diabetes Association, 40% of Qatari kids suffer from diabetes and the cause is mainly life style. Onigiri is a technology-based solution that aims at educating teens & kids about the side effects of Diabetes through a game. It is well known that teenagers are addicted to electronic games. Using something that is close to their universe facilitates reaching out to them. The solution is based on the idea that ‘prevention is better than cure’. Creating awareness about the disease may slow its progression. This solution is educational but also recreational, as it is learning through play. Onigiri is a game that can be played on iPod and iPhone. ‘Onigiri’, the game character’s name means ‘Rice ball’ in Japanese. He looks like a round shaped grain of rice. Onigiri is an overweight pet that is affected by diabetes and needs to be taken care of. The game player will be helping Onigiri to exercise and eat healthy food to keep its blood sugar level acceptable. The player will be giving Onigiri its insulin shots when needed and help it sleep when needed. The game uses the iPod’s accelerometer technology for some functions of the game. Using this technology, the player can make Onigiri automatically exercise if he does too. For example, when the player is jogging and the Onigiri application is active, Onigiri will automatically exercise due to the iPod’s accelerometer. Through the game, the player learns how the life of a diabetic patient looks like and can understand the proper ways of managing Diabetes by having a balanced diet and a healthy life.

54



StudentTRAK

Poster Q26

Authors Amna Al-Hitmi (IS 2011) Mariam Al-Sunaidi (CS 2011) Douaa Dalle (IS 2011) Yasser Khan (IS 2011) Faculty advisor Ian N Lacey, Ph.D. Category Information Systems Abstract We present details of the development of a student payment system called StudentTRAK for Carnegie Mellon Qatar’s employed students. This is intended to replace the current a paper-based system where the students have to fill in a time sheet each month for each job and then collect the signature of his or her supervisor. This system will serve as the central online hub for monitoring and tracking all forms of student employment, so that there is no discrepancy with regard to work hours regardless of department. The current system is very time consuming in terms of the validation process. It has a high error rate in terms of data entry; it is difficult to manage in terms of access to student/job records; and it causes latency in student payment. StudentTRAK will speed up the current process, make the timesheet validation process easier and reduce the error rate. Development of the StudentTRAK system was based upon interviews with currently employed students, a student supervisor and a student counselor. To build the system, we designed the database first and then we used our designs to guide us through the programming part. We are using web tutorials and books to help us finish a running prototype of the final system. StudentTRAK is an on-line system where students, supervisors and counselors can log in and perform a lot of functions electronically. For a student, the functions include choosing one of his or her jobs, filling an interactive timesheet that can be edited very easily and sending the timesheet electronically to his or her supervisor for approval. For a supervisor, the functions include creating a new student job, enrolling a student in a job, receiving a student timesheet for approval, viewing the timesheet, rejecting the timesheet or approving it and forwarding it to student counselor. For a student counselor, the functions include receiving the approved timesheet from supervisors, approving the timesheet and generating a PDF copy for further use or rejecting it.

56



Transactional Referral System

Poster Q27

Authors Amal Al Barwani (IS 2010) Khalid Alsooj (IS 2010) Shakir Hussain (IS 2010) Nasreen Zahan (IS 2010) Faculty advisor Divakaran Liginlal, Ph.D. Category Information Systems Abstract Transactional Referral System (TRS) is a web-based application that enables two-way communication between doctors at various clinics in Qatar and specialists at the Qatar Diabetes Association (QDA) with regard to referring diabetic patients. During the team’s introductory meetings with the project’s primary stakeholders –Qatar Diabetes Association and ictQatar, the Supreme Council of Information and Communication Technology - we were informed that the existing patient referral process doctors and QDA specialists use contains several problems. First, the existing system is paper-based. This creates two reliability issues. The referrals may get lost while the patient transports it from the clinic to the QDA. And it takes the doctors and QDA specialists additional time and effort to locate prior documentation. The second concern is giving the doctors feedback through a phone call is not reliable as the doctor may not be in, or busy with a patient or on the phone. The TRS enhances communication between doctors at health care centers and QDA specialists. It enables two-way communication between these two parties. The QDA specialists will be able to receive electronic referrals of diabetic patients from doctors at the health care centers, which is a significant improvement over the current inefficient paper-based system. They will also be able to send their feedback back through TRS, thus making the communication process more effective. Since the doctors have very limited time to diagnose and treat the patient, a lot of effort has been put into the usability of TRS. The doctors will be able to fill out the forms and submit the referrals in a very limited time, thus reducing the overall time spent on the system and more time spent with the patients.

58



60


Post-graduate posters


Assistive Educational Technology Project

Poster Q28

Authors Sarah Belousov Hend Gedawy (CS 2009) Ermine Teves Faculty advisor M. Bernardine Dias, Ph.D. Category Post Graduate Abstract The Assistive Educational Technology project explores the role that computing technology can play in developing educational tools for both visually-impaired and deaf communities in Qatar and around the world. Faculty, research staff and students at Carnegie Mellon University in Doha and Pittsburgh are currently developing and field testing automated tutors and educational games to enhance the lives of these populations. Braille is the primary method of reading and writing for the visually-impaired, while Sign Language is the main method of communication for the deaf. Despite the importance of literacy to employment, social well being and health, the literacy rate of the visuallyimpaired and deaf populations around the world is estimated to be quite low. The significant educational challenge these populations face is that they are a minority and as a result have fewer opportunities to reinforce their literacy and communication skills, and fewer educational tools targeted to their needs. Technology tools such as automated tutors and educational games thus have the potential to significantly enhance their social and economic well being. The pioneering tools developed in this project include the Braille Writing Tutor that supports Arabic, English and other languages; DeSign, a computer program that teaches Arabic and English Sign Language; and a mobile phone application that targets English literacy. These interactive tools make use of games and other approaches to enhance the tutoring experience and are continually being enhanced through field testing with our partners in Doha, Pittsburgh and other communities.

62



Coverage Problem in Wireless Multimedia Sensor Networks

Poster Q29

Author Vikram P. Munishwar Faculty advisor Nael B. Abu-Ghazaleh, Ph.D. Category Post Graduate Abstract Smart camera networks are becoming increasingly popular as a means for providing visual coverage of places of interest. Although, the traditional surveillance systems consist of multiple cameras, they need a human-in-the-loop to control and monitor the video being collected by the cameras. Removing the dependence on human control of the cameras significantly lowers the overhead of deploying such monitoring. In this project, we consider the problem of automatic control of the cameras to maximize coverage of a set of targets they are monitoring. Specifically, we present an optimization problem formulation with the goal of maximizing the number of targets covered. Since this problem is an NP-hard, we propose a computationally efficient heuristic to obtain a near-optimal solution to the problem. Next, we consider the case of mobile targets, where the cameras have to maintain closeto-optimal coverage as the targets move. While centralized solutions work well for static targets, the configuration of cameras seize to be optimal as the targets move. In order to address this problem, we propose semi-central approaches for coverage optimization. We also provide a collaboration mechanism among neighboring cameras in order to hand over the responsibility to cover a target as it is moving. Finally, we consider the problem of scalability due to the semi-centralized nature of the proposed mechanisms, and propose a novel hierarchical coverage management approach.

64



Hala the Bilingual Robot Receptionist

Poster Q30

Authors Ameer Abdulsalaam (Research Engineer) Imran Fanaswala (Research Programmer) Wael Ghazzawi (Research Engineer) Faculty advisors Brett Browning, Ph.D. Majd Sakr, Ph.D. Category Post Graduate Abstract As robots become integrated into industry, commercial and personal settings, they will need to interact effectively with people. In this project, we are investigating techniques to enable effective human robot interaction (HRI) by exploring the nature of interactions in a social, multi-cultural and multi-lingual setting. We have developed Hala, a bilingual, pan-Arabic robot receptionist as a mechanism to study HRI in such a setting. Hala has been deployed for the last 6-months in the main reception area of the Carnegie Mellon Qatar Building in Education City, where she interacts with visitors, students, staff and faculty in either English or Arabic. She responds to queries related to campus directions, weather, time, local events, local organizations, her personal life and her immediate surroundings. Hala’s user experience is centered around three things: invitation, how do we initiate a conversation; interaction, how do we provide relevant answers; and engagement, how do we keep the user engaged for a prolonged period. Future work involves examining and improving these three key areas through anonymized logs of user behavior.

66



Initial Findings on Provisioning Variation in Cloud Computing

Poster Q31

Author Suhail Rehman (Research Analyst, Cloud Computing Lab) Faculty advisor Majd Sakr, Ph.D. Category Post Graduate Abstract Cloud computing offers a paradigm shift in management of computing resources for large applications. It has the potential to offer scientific researchers compute resources on demand on a pay-as-you-go basis, eliminating the need for upfront investment and management of large, expensive high performance computing (HPC) systems. Using the Infrastructure-as-a-service (IaaS) cloud computing model, researchers today can request provisioned virtualized resources such as CPU, memory, disk, and network access. These resources can be scaled on demand as well as released when not in use. While the behavior of scientific applications on traditional HPC hardware is understood, in case of dynamically provisioned hardware in IaaS systems, it is not. This is one the barriers to increased adoption of cloud computing technologies in HPC. On a cloud, the performance of an application may vary according to the virtual configuration of the provisioned resources. This encompasses the number of virtual machines (VMs) provisioned and each VM’s configuration in CPU power and memory allocation. However, on a cloud system, all of these parameters can be satisfied using different physical configurations. We conjecture that variations in mapping VMs to physical resources can have a significant impact on performance. This mapping is invisible to the cloud user. The impact on performance is due to the provisioning variation of inter-VM latency and bandwidth. In this work, we study the effects of this “provisioning variation” on the performance of a few toy applications implemented in MapReduce executed on the cloud using the Hadoop infrastructure. In future work, we plan on extending this study to scientific applications so that we can develop a better understanding of their requirements for deterministic performance on the cloud. This work will then guide the migration of scientific applications from HPC systems to cloud computing systems.

68


Suhail Rehman

Initial Results of Provisioning Variation in Cloud Computing

IaaS

PaaS

Research Analyst - CMUQ

Prof. Majd Sakr

Associate Teaching Professor

SaaS

Application Layer Application Platform Layer

Infrastructure Layer

App

App OS Hardware

App

Virtualized Stack

Traditional Stack

Software Layer

App

App

App

OS

OS

OS

Hypervisor Hardware

Hardware Layer

= Virtualized Resource

map

reduce

QLOUD

hardware

VM

VM

Physical Host

VM

Physical Host

software

VM

Physical Host

Physical Host

4 VMs across 4 Nodes VM

VM

Physical Host

VM

VM

Physical Host

4 VMs across 2 Nodes

VM

VM

VM

VM

Physical Host

4 VMs across 1 Node

Cloud Computing Lab


Interference in Multihop Wireless Networks

Poster Q32

Author Saquib Razak, Ph.D. Faculty advisor Nael B. Abu-Ghazaleh, Ph.D. Category Post Graduate Abstract Wireless networking is a critical technology that plays a central role for providing connectivity to mobile data and voice users. Multi-Hop Wireless Networks (MHWNs), which include mesh, sensor and ad hoc networks, are forecast to play an important role in an Internet that will grow increasingly wireless at the edge. Interference plays a complex and often defining role on the overall performance of wireless networks, especially in multi-hop scenarios. Understanding this role is critical for estimating the performance of these networks, and in turn for developing effective protocols for them. In the presence of interference, Carrier Sense Multiple Access (CSMA) Medium Access Control (MAC) protocols are known to suffer from the hidden terminal and exposed terminal problems, which can cause poor performance and unfairness. Depending on the relative location of interfering sources and destinations, several modes of interference exhibiting different behavior occur. In this work, we start with categorizing these interference modes between two concurrent transmissions (the basic unit of interference in a network) in order to understand their behavior. We term these modes of interference as interference interactions. We identify five types of interactions. We develop closed form geometric models to determine occurrence frequencies for each category and verify these models against Monte Carlo simulation based results. Our next step is to extend this work by using a more realistic model for packet reception. We use Signal to Interference and Noise (SINR) model and capture effects for packet reception that allows for a more accurate modeling of packet reception. Based on this model, which makes our analysis more complicated, our interaction categories increase to 11. We extend this analysis to study how TCP connections, which involve bidirectional flows, behave over wireless chains. First, we break down and examine the types of chains that occur most frequently in TCP configurations and classify them by the nature of the MAC level interactions that arise in each. We then show that the throughput of TCP over a wireless chain is greatly affected by the type of interactions within the chain. We discuss the implications of the MAC level interactions on network performance: specifically, route instability and number of retransmissions. 70



LNG Pipe Vision

Poster Q33

Authors Hatem Alismail (CS postgraduate) Brett Browning, Ph.D. (CS faculty) Peter Hansen (CS postdoc) Peter Rander, Ph.D. (CS faculty) Faculty advisor Brett Browning, Ph.D. Category Post Graduate Abstract Liquid Natural Gas (LNG) processing facilities contain large networks of pipes, and the inspection of these pipes for corrosion is critical for safety. One method used to detect corrosion is to use magnetic flux leakage sensors, for example, to measure (monitor) changes in pipe wall thickness over time. However, a limitation of this approach is that only one source of data is obtained, namely pipe wall thickness, and information regarding the internal appearance of the pipes, which can also be used to detect potential points of corrosion, is not obtained. We propose that a vision system equipped on a pipe-crawling robot is an ideal candidate for pipe inspection as it can be used to produce accurate appearance maps, which are stitched images of the interior surface associated with dense 3D measurements of pipe wall radius. We present preliminary work that focuses on the former; to produce accurate stitched images of the interior surface from monocular image sequences captured in a pipe. This requires, for every image in a sequence, the pose of the camera relative to the pipe to be known. We investigate two methods for measuring the change in appearance between images that are in use to estimate camera pose. The first is a dense algorithm that uses the intensity values of all pixels in adjacent images. The second is a sparse algorithm that uses salient point feature correspondences between adjacent images. Experimental results for both methods are presented for multiple images sequences captured in two pipes - one is 4 meters long and 16� in diameter, the other is 6 meters long and 6� in diameter. These results show that accurate pose estimates can be obtained that consistently have errors less than one percent for distance traveled down the pipe. Examples of the stitched images are also presented that highlight the accuracy of these pose estimates.

72



Open Domain Named Entity Recognition for Arabic

Poster Q34

Author Behrang Mohit (post doctorate) Category Post Graduate Abstract We report our research progress on Named Entity Recognition (NER) from Arabic Wikipedia text. Automatic recognition of named entities, such as person, location and organization, generally relies on statistical learning methods. These methods require large volumes of labeled training data. For the Arabic language, the labeled data is scarce and limited to political news domain. Working on an open domain text like Wikipedia requires major modification to both the problem definition and learning framework. For example, the basic three classes - person, location and organization - are not enough to cover named entities. Also NER models, which are trained on the narrow political domains, are not robust enough in other domains. Here we present two dimensions of our ongoing work: (1) Modification of the named entity labeling from a closed class framework to a more relaxed one where annotators introduce new named entity classes based on the genres of the Wikipedia article. (2) Usage of the NER model that is trained on a narrow domain to find named entity boundaries from the open-domain Wikipedia text. Moreover, we motivate our future work on using Wikipedia’s hierarchical and multilingual structure to enrich the baseline NER models. This work is part of the American and Qatari Modeling of Arabic (AQMAR) project that is funded by QNRF through the NPRP fund (08-485-1-083).

74



Student Modeling for the Adaptive Braille Writing Tutor

Poster Q35

Author Hend Gedawy (CS 2009) Faculty advisors Brett Browning, Ph.D. M. Bernardine Dias, Ph.D. Category Post Graduate Abstract Braille is a system of reading and writing in which patterns of raised dots are used to represent letters and words in a given language. These raised dots can be read using the fingers, thus enabling blind people to be literate. The traditional tool used to write Braille is the slate and stylus. However, blind children learning how to write Braille using this tool face several challenges including the physical and mental exertion required, and the delayed feedback on what is written. The Adaptive Braille Writing Tutor was invented by researchers in the TechBridgeWorld organization at Carnegie Mellon as an affordable and entertaining technology to make it easier for blind children to learn to write Braille. It is an interactive tutoring system that gives the user immediate audio feedback. In this work, we enhance the intelligence of our Braille Writing Tutor by enabling it to accurately model the student’s knowledge and thus tailor the tutoring to fit the learning needs of the user. After a comprehensive exploration of various approaches, we adopted the Bayesian Knowledge Tracing approach for our work. This method has been used in several tutoring systems, and has shown improvement over alternatives, in assessing the student’s knowledge. It uses Bayesian networks for longterm assessment of student knowledge, plan recognition and prediction of the student’s actions during problem solving. Being able to accurately model student knowledge will enable the Braille Writing Tutor to adjust its tutoring decisions, such as giving hints and or changing the difficulty level, according to the student’s needs, with the ultimate goal of enhancing its educational impact.

76



Task Allocation and Coordination in a Disaster Response Scenario

Poster Q36

Authors Ameer Abdulsalaam (Research Engineer) Imran Fanaswala (Research Programmer) Faculty advisor Bernardine Dias, Ph.D. Category Post Graduate Abstract The vision of humans and robots working together as a team to execute complex tasks has motivated research in efficient team coordination and task allocation. Complex and dynamic domains, such as disaster response, present challenging scenarios where future teams of humans, robots and software agents must coordinate under a variety of uncertain conditions with imperfect information to carry out critical missions with limited resources. An important aspect of effective team coordination in these settings is an efficient allocation of tasks to appropriate team members that takes into account task and team constraints. In this work we explore three topics relevant to the team coordination problem in disaster response: (1) Efficient coordination of large teams operating in dynamic and uncertain settings, (2) Optimal planning for small teams performing critical tasks, and (3) Effective task allocation in human-robot teams. We employ the TraderBots market-based approach to efficiently coordinate larger teams in dynamic settings. Optimal planning for smaller teams is accomplished through a branch-and-price mathematical programming algorithm. Finally, we develop a mobile phone application to accomplish localization and task allocation to human team members in a human-robot team setting.

78



Task Allocation and Scheduling for Constrained Teams

Poster Q37

Author G. Ayorkor Korsah Faculty advisor M. Bernardine Dias, Ph.D. Category Post Graduate Abstract In many domains, such as disaster response, agriculture, construction and mining, teams of robots and humans can work together to accomplish a collection of spatially distributed tasks. Each agent - robotic or human - has specific capabilities that make it particularly suited to certain tasks, and the problem of coordinating a team of robots and humans to achieve the desired goals is an important one. Key questions to answer in solving the coordination problem are: which agents should perform which tasks, in what order, and when? This research focuses on computing the optimal task allocation and schedule for such coordination problems in which there are several real-world constraints. For example, while some tasks are independent of each other, other tasks may be related by precedence, simultaneity or proximity constraints. These constraints are in some cases a result of the complementary capabilities of robots and humans that require them to cooperate to achieve certain goals. Many tasks may have a fixed pre-specified location at which they must be performed, but others may require movement from one location to another, and yet others may have a choice of locations where they may take place. We apply mathematical programming techniques to this complex multi-agent task allocation and scheduling problem. Although the proposed mathematical programming approach enables finding the optimal allocation and schedule, the approach is not suited to dynamic situations where new tasks are coming in over time, or the environment is changing in some manner. We illustrate how the mathematical programming approach can be combined with a marketbased approach to handle such dynamic situations.

80



Touch-Screen Interface Design for Multicultural Settings

Poster Q38

Author Ameer Abdulsalem Faculty advisor Majd Sakr, PhD. Category Post Graduate Abstract Our work focuses on developing multi-lingual intuitive user interfaces for a one-time, non-expert system user. Effective interfaces that meet these criteria have to be intuitive, user-friendly and natural in guiding the one-time-user through the system interface. Such systems require easy input methods and easy to navigate interfaces. Human-Computer Interaction systems typically require the user to input data to a computer system. Input systems can span the spectrum of free-form natural language to constrained keyword-specific. Although free-form natural language is easier for the user it is very hard to build systems to extract meaning from the natural language input. On the other hand, constrained keyword-input is easy to parse and process but requires user training and experience before it is mastered; this approach is not suitable for the one-time user. A middle ground approach involves the use of an abstraction layer that provides a set of generalized keywords that the user selects from in order to convey input to the computer system. These keywords effectively provide categorization that encompasses all possible queries. In our work, we attempt to develop an input system that capitalizes on the human’s abilities and logic to correctly form an input query from the provided generalized choices. To evaluate this approach, we utilize an information retrieval system for an organization of fixed structure and function. We use this test-bed to evaluate the efficacy of using this abstraction model to develop intuitive user interfaces that enable a one-time-system-user to effectively navigate and reach the desired information. Specifically, we have designed and developed an intuitive touch-screen interface for an information retrieval system using Carnegie Mellon Qatar as the target organization. The multi-lingual interface allows us to evaluate this method across language and cultural variations. To aid the user, a language-specific animated face acts as a voice-guide to aid navigation through the interface. In this work we present an overview of the data abstraction and interface design that went into this system. Furthermore we present some initial insight gained through preliminary experiments to compare the Arabic and English interfaces factoring in the effectiveness of the animated face at enhancing the user experience. 82



Wireless Network Optimization Using Software-Defined Radios Poster Q39

Author Vinay Kolar (CS Post-doc) Faculty advisor Nael B. Abu-Ghazaleh , Ph.D. Category Post Graduate Abstract Wireless networks are corner-stones for providing instant access to information in this rapidly mobile world. They provide seamless anytime-anywhere connectivity to mobile devices such as smartphones, cameras and laptops. Smart “pocket” devices, such as the iPhone, are capable of connecting to the Internet virtually from anywhere through WiFi and 3G access. These gadgets, coupled with the increasing user’s thirst to constantly access media-rich content such as YouTube, have not only increased the importance of wireless networks, but have also raised significant challenges. Catering huge capacity demands in wireless networks is a hard problem. Achieving higher capacity in wireless network - unlike the wired network - is challenging due to sparse bandwidth and errorprone wireless channel. Optimizing such a resource-constrained and dynamic network provides poses hard problems. In this poster we discuss our solutions to a few of the fundamental problems that improve available capacity to the end-user by: (1) dynamically recognizing the network topology, (2) identifying higher capacity routes, and (3) scheduling transmissions such that there are no errors. Current off-the-shelf wireless devices are incapable of - or do not allow - accessing lower-level details, such as wireless channel quality, that provide valuable control knobs to achieve high-capacity. Hence, we demonstrate using Software-Defined Radios (SDRs), which are advanced programmable radios, to extract and exploit lower layer information. The poster has two main contributions: we develop a Network Monitoring Tool to visualize the intricate details of wireless networks, and a suite of efficient protocols to improve the available capacity. The “Network Monitoring Tool” dynamically identifies the nodes in the network and analyzes the transmission pattern across the network. The protocol suite includes measurement protocols and high-capacity routing protocol. Measurement protocols probe the nodes to identify the available capacity at each node. Our routing protocol chooses a set of nodes to route a packet from source to end-destination such that the end-capacity for the connection is maximized. The Network Monitoring Tool and Protocol Suite adapt to dynamically changing network. The measurement tool is extremely useful for analyzing the causes of inefficiency in wireless networks. Moreover, the framework of measurement protocols are highly extensible and useful for developing higher layer protocols. For example, applications such as video streaming can utilize the measured information to intelligently tune the video codec based on the available capacity. We believe that the tool and protocol suite provide a fundamental understanding of detailed wireless network behavior, and a realistic proof-of-concept solution to optimize lower-layers in wireless networks. 84



P.O.Box 24866 | Doha, Qatar | Ph: +974 454 8400 | Fax: +974 454 8410 | www.qatar.cmu.edu


Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.