MS Thesis Defense

Distributed Model Consensus for Models of Locally
Biased Measurements in Wireless Sensor Networks

Jacob Thompson

4:00pm Wednesday, 25 April 2012, ITE 325B, UMBC

Wireless sensor networks (WSNs) consist of interconnected microsensors, each of which collects measurements from its local environment, which are often used in monitoring and control applications. These applications make inferences about the global and local states of the deployment environment. However, due to the limited communication and energy resources at the sensors, gathering all the raw data at a central fusion/control point is impractical.

Hence, it is essential to have distributed learning and inference in WSNs, such as learning a consensus model from the locally learned models. Consensus is challenging due to the limited resources of the sensors and the inherent bias of the individual sensor models learned from their local sensing environments. Two leading approaches for this problem are the approach by Zheng et al which uses loopy belief propagation on a certain graphical model based on the WSN topology and the local models, and the approach by Xiao et al which relies on gossip averaging of the parameters of the local models.

We focus on multivariate linear regression models, such as Bayesian, Ridge, and Lasso regression models. We analyze and extend the loopy Gaussian belief propagation (GaBP) approach to model consensus, and compare its performance to the gossip averaging approach. We experimentally find that GaBP tends to converge much faster than gossip averaging, but to a less accurate estimated consensus model (especially in the presence of multiple cycles in its corresponding graphical model). We also find that gossip averaging along paths in the WSN, tends to provide much faster convergence to more accurate estimated consensus models as compared to GaBP.

Committee: Professors Kostas Kalpakis (chair), Tim Oates and Yun Peng

photo courtesy www.umbc.edu/gsa

More than two-dozen Computer Science and Electrical Engineering graduate students are slated to present at the Graduate Student Association's annual Graduate Research Conference (GRC). The GRC will be held this Friday, April 27 and features a keynote presentation from Dr. Michael I. Nishimura, a UMBC alumnus who now works at Loyola University Chicago as a professor in the Department of Surgery, the Associate Director of the Oncology institute, the Associate Director of the Cancer Center Translations Research, and as Program Director of Immunologic Therapeutics.

This year, the CSEE presence at the GRC is even stronger than it was last year, when three CSEE students were awarded for outstanding presentations. The honorees were Varish Mulwad who received an award for an outstanding oral presentation on his dissertation research entitled "Generating Knowledge from Tables," Kavita Krishnaswamy who received an award for an outstanding oral presentation on her thesis research entitled "Path planning a roboticarm efficiently," and Akshya Iyengar who received an award for an outstanding poster presentation for her thesis research entitled "Estimating Temporal Boundaries of Events using Social Media Data".

Take a look at the list below, which includes the 27 CSEE graduate students who will present at the GRC this year.

Check out some photos from last year's conference.

Click here for a printable list of CSEE presenters.

Oral Presentations
Location: Sondheim 203

“Noise Reduction in AIRS Infared Earth Observing Radiance Grids using the Obscov Algorithm”
David Chapman, Milton Halem, Phuong Nguyen, and Jeff Avery
Time: 9:00 a.m.

“Heart Disease Prediction Model: A Data Mining Approach”
Soma Das
Time: 9:15 a.m.

“Measuring the Pulse Duration of Ultrafast Lasers”
Jared Dixon
Time: 9:30 a.m.

“Using Deceptive Packets to Increase Base-Station Anonymity in Wireless Sensor Network”
Yousef Ebrahimi and Mohamed Younis
Time: 9:45 a.m.

“SmartRate: A Rating Interpretation Mechanism for Agents in Smart Grid Markets”
Yasaman Haghpanah
Time: 10:00 a.m.

“Visualizing Changes to Directory Structure to Support Digital Forensics”
Timothy Leschke
Time: 10:15 a.m.

“Extracting Semantic Linked Data from Tables”
Varish Mulwad
Time: 10:30 a.m.

“Automatic Identification of Prescription Drugs Using Shape, Imprint, and Color”
Adrian Rosebrock
Time: 11:00 a.m.

“An Optical Sensor for a Ceramic Water Filtration System for the Detection of E.Coli Using a Microfluidic Chip”
Serina Woods
Time: 11:15 a.m.

Poster Presentations
Location: Albin O. Kuhn Library: 7^th Floor

“Witness-based Saboteur Detection in Multi-agent Systems”
Petr Babkin
Time: 11:00 a.m.

“Entity Linking and Disambiguation for Smartphone Platforms”
Anurag Korde
Time: 11:00 a.m.

“Third-order Quadratically Converging, Quasi-Newton Optimization”
Rory Mulvaney
Time: 11:00 a.m.

“Supervised Learning Techniques for Predicting Risk of Breast Cancer using Genetic Information”
Aniket Bochare
Time: 11:12 a.m.

“Prostate Cancer Prognosis using Genomic Data”
Rohit Kugaonkar
Time: 11:12 a.m.

“Using Supervised Techniques for Classification of Conventional Data Items”
Nikhil Puranik
Time: 11:12 a.m.

“Surface Emitting Quantum Cascade Laser Array”
Xing Chen, Liwei Cheng, Dingkai Guo, and Fow-Sen Choa
Time: 11:24 a.m.

“Learning Sensitive to Multiple Sources of Costs”
Zachary Kurtz
Time: 11:24 a.m.

“Exploring Hidden Markov Model for Semantic Activity Prediction”
Amey Sane
Time: 11: 24 a.m.

“Calculating Representatives of Geographic Sites across the World”
Ashwinkumar Ganesan
Time: 11:36 a.m.

“Older Adults Interactions with a Touch Table Top Display Space”
Galina Madjaroff
Time: 11:36 a.m.

“Rendering of Smoke and Fire in a 3D Volume with Multiple Scattering”
Taekyu Shin
Time: 11:36 a.m.

“Chronic Disease Prediction: An Experimental Analysis Using the K-nearest Neighbor Algorithm”
Matthew Gately
Time: 11:48 a.m.

“Situation Aware Intrusion Detection Model”
Sumit S. More
Time: 11:48 a.m.

“Unsupervised Coreference Resolution for FOAF Instances”
Jennifer Sleeman
Time: 11:48 a.m.

“Towards an Intuitive Query System for DBpedia”
Lushan Han
Time: 12:00 p.m.

“Link Prediction Using Frequent Subgraphs”
Maksym Morawski
Time: 12:00 p.m.

“Modeling Motheye Antireflective Structures for Increased Coupling through As2S3 Optical Fibers”
Robert J. Weiblen
Time: 12:00 p.m.

Classes each Wednesday evening starting October 17.

This five week workshop prepares participants to take the computer-adaptive Graduate Records Examination. The purpose of the course is to improve students' performance on the GRE by:

• Providing students with successful test management techniques
• Familiarizing students with question types specific to the GRE
• Teaching the logical, rhetorical, and mathematical problem-solving techniques relevant to the GRE
• Teaching essential grammatical and mathematical concepts and techniques in application to real test problems
• Identifying the specific problems posed by the GRE, and analyzing common test-taking errors
• Improving performance and reducing test anxiety through practice testing

This workshop is taught by Dr. Anthony Poplawski, who has helped thousands of students perform better on their GRE examinations. While this workshop provides participants with very valuable information, techniques and strategies for maximizing their potential on the GRE, please note that it does not replace studying. In addition to taking this workshop we recommend that students dedicate significant time to studying for the examination in order to be truly prepared.

Please visit the Graduate Record Examination web site at www.gre.org to view information on the GRE exam.

TO REGISTER AND LEARN MORE ABOUT THE COURSE, PLEASE VISIT:

http://www.umbc.edu/testprep/gre.html

Take advantage of early bird discounts by registering before September 26, 2012!

Come to UMBC and prepare for the types of questions included in the Law School Admissions Test. Designed to improve performance on the competitive LSAT, this intensive course focuses on:

1. Test-taking techniques
2. Time-saving methods
3. Improving logical reasoning
4. Practicing with sample tests
5. Enhancing reading comprehension
6. Reducing test anxiety

Taught by Dr. Anthony Poplawski of Spiegelberg and Associates, UMBC’s Test Preparation Courses are acclaimed for their comprehensiveness, revealing the true character of the test, their focus on maximizing scores, their intellectual seriousness, and reducing the test candidate’s test anxiety. Instruction is at the highest professional level by experienced college faculty.

Please visit The Law School Admission Council at www.lsac.org to view information on the LSAT exam.

TO REGISTER AND LEARN MORE ABOUT THE COURSE, PLEASE VISIT:

http://www.umbc.edu/testprep

Analyzing Social Media Data
Morgan Madeira
Junior, Computer Science

Social media has increasingly become an outlet for expression for a large part of our society. Literature suggests that analyzing data from these sites can lead to improvements in areas such as health-care and search-ad targeting. Users of these sites often associate with many other users described as “friends,” even if they do not have a strong connection, or what would be described as friendship in daily life. It is valuable to determine the strength of relationships between users and to identify communities within social networks. These communities represent people with similar characteristics, which are used by applications to solve many real-world problems. For instance, it is useful to identify groups that are interested in a specific movie genre. Information about these groups can be used to target movie advertisements towards the people most interested in that genre. These types of problems have similar characteristics to identifying close friends. We have created a system to collect and analyze the data about user characteristics, while being respectful of privacy concerns. The system is composed of a front end Facebook application and a back end machine-learning based tool. The front end component gathers data about a user and their friends. The back end uses the collected data and machine-learning techniques to determine relationships between users.

To learn more about the project, check out an interview with Morgan.

Catch Morgan's poster presentation at URCAD this Wednesday, April 25 in the University Center Ballroom from 10:00 a.m. to 12:30 p.m.

Multiclass Datasets, Their Predictions, and Their Visualization
Wallace Brown and Alexander Morrow with Kevin Winner
Senior, Computer Science
Sophomore, Computer Science

Many datasets contain a wealth of information. For example, a person may be described by their race, age, gender, income, marital status, nationality, level of education, etc. By analyzing this data, we can form educated and accurate predictions about individuals. We can, for instance, determine that a person with a particular race, age, nationality, and income is likely to be a college undergraduate. Our goal is to develop ways to visualize these predictions and the uncertainty associated with the predictions. Displaying data in a scatterplot is a standard means of describing two-dimensional information. However, displaying high-dimensional data (i.e., data that includes many attributes, such as age, race, and income) is significantly more challenging.  We present a means of visualizing high-dimensional data sets and the predictive models derived from the data, using existing dimension reduction techniques and novel glyph-based displays.

To learn more about the project, check out an interview with Wallace and Alexander.

Catch Wallace and Alexander's presentation at URCAD this Wednesday, April 25 in the University Center 312 from 1:15 to 1:30 p.m.