MS Thesis Defense

distratto: Real-time Impaired Driving Detection Using Multiple Textile and Inertial Sensors

Tsu An Chen

1:00-3:00pm Tuesday, 23 December 2014, ITE 341

Statistical data shows that driving-related accidents and human casualties caused by vehicles are on the rise in the US and globally. Most of these accidents are cause by impaired or distracted driving. Existing systems that detect impaired driving use cameras that perform eye and head tracking and do not capture full-body movements that are indicative of dangerous driving. To address this problem, in this thesis we present a system, distratto, that uses capacitive textile sensors embedded into car seats, headrests, and arm rests to capture whole body motion, and inertial and GPS sensors for determining vehicle speed and turns. Using a combination of these sensors and a tiered signal processing algorithm, we infer attributes that are indicative of impaired driving. We have developed a fully functional prototype of distratto that we evaluate in a real vehicle setting. We show that our system can detect impaired driving instances and driver movements with high accuracy.

Committee: Drs. Nilanjan Banerjee (chair), Ryan Robucci, Chintan Patel

Ph.D. thesis defense

Paper Form Classification for Information
Systems Strengthening in Developing Countries

Huguens Jean

1:00pm Friday, 19 December 2015, ITE 325b

In developing countries, people are now more likely to have access to a mobile phone than clean water, making cellular based technology the only viable medium for collecting, aggregating, and communicating local data so that it can be turned into useful information. While mobile phones have found broad application in reporting health, financial and environmental data, many data collection methods still suffer from delays, inefficiency and difficulties maintaining quality. In environments with insufficient IT support and infrastructure, and among populations with limited education and experience with technology, paper forms rather than electronic methods remain the predominant means for data collection.

To meet the digitization needs of paper driven data collection practices in developing countries, SHREDDR proposes an end-to-end architecture that transforms paper form images into structured digital information on-demand. To facilitate the automatic extraction of input regions in form images, this thesis extends the SHREDDR architecture with the necessary capabilities to efficiently classify form images according to their template document. Specifically, it introduces a novel framework for visually identifying form templates by decomposing the template identification problem into three distinct tasks: retrieval, learning and matching (RLM).

Given a query form instance, the retrieval component finds and ranks the topmost h similar templates. If h>1, the matching component uses full image registration to conduct a more rigorous assessment of the visual similarity between the query form instance and the candidate templates. After matching, the retrieval’s preliminary ranking is adjusted, if necessary. The topmost candidate template with the highest registration score satisfying a global alignment threshold denotes the input form’s template. Based on the answer obtained from matching, the learning component updates the retrieval so that it can provide a better ranking in future searches. If h=1, the RLM bypasses matching and uses the retrieved template as the final classification.

Based on the proposed framework, the present thesis investigates form classification under the conditions of known and unknown template classes. A pilot study integrating the RLM into the SHREDDR system demonstrates its classification accuracy and its impact on digitization efficiency.

Committee: Drs. Timothy Oates (Chair), Fow-Sen Choa, Janet Rutledge, Jesus Caban, Nilanjan Banerjee

MS Thesis Defense

SAX-BOP: Epileptic Seizure Detection using
Symbolic Aggregate Approximation and Bag of Patterns

Sidharth Allani

1:00pm Friday, 12 December 2014, ITE 325b

Epilepsy is a chronic neurological disorder that makes patients susceptible to experiencing recurrent seizures. A seizure occurs when abnormal activity in the brain leads to involuntary body moment, lack of awareness or behavior, short-term loss of memory or attention, short-term unconsciousness, or body convulsions. Epilepsy affects three million people in the United States and accounts for $15.5 billion in direct and indirect costs.

Epilepsy has many different causes, and often no definite cause can be found. Patients who suffer from intractable seizures experience unpredictable and frequent seizures that cannot be controlled using anti-seizure drugs. Such seizures leave the patient traumatized and, due to their uncertainty, the patient’s mobility and independence are restricted, resulting in social isolation and economic hardship.

The research in this thesis aims to detect epileptic seizures and to analyze the performance of Symbolic Aggregate approXimation and the Bag of Patterns representation for seizure event detection. We use Electroencephalogram (EEG) recordings as the data source for seizure detection, which is the recording of electrical activity along the scalp that measures ionic current flows within the neurons of the brain. These signals are a good source of information about abnormal activity in the brain and are helpful in the process of epileptic seizure detection. This problem becomes challenging because of the enormous size of the EEG data, making it difficult to effectively and efficiently analyze these signals and detect a seizure. We use Symbolic Aggregate approXimation (SAX) and the Bag of Patterns Representation (BOP) and analyze their performance with EEG time series data to detect seizures.

Committee: Drs. Tim Oates (chair), Tim Finin and Tinoosh Mohsenin

Ph.D. Dissertation Proposal

Learning Representation and Modeling Time Series

 Zhiguang Wang

10:00-12:00 Friday, 12 December 2015, ITE 325B

Most real-world data has a temporal component, whether it is measurements of natural (weather, sound) or man-made (stock market, robotics) phenomena. Analysis of time-series data has been the subject of active research for decades and is still considered to be a challenge in machine learning and data mining due to the properties of temporal data.

Traditional approaches for modeling and representing time-series data fall into three categories. Non-data adaptive models, such as Discrete Fourier Transformation (DFT), Discrete Wavelet Transformation (DWT), and Discrete Cosine Transformation (DCT), compute the transformation with an algorithm that is invariant with respect to the data. Data adaptive approaches such as Symbolic Aggregation approXimation (SAX), Piecewise Linear Aggregation (PLA), and shapelets compute transforms that are highly dependent on the data. In model-based approaches such as AutoRegressive Moving Average models (ARMA), Linear Dynamical Systems (LDS), and Hidden Markov Models (HMMs), the underlying data is assumed to fit a specific type of model. The estimated parameters can then be used as features in, for example, a classifier.

However, more complex, high-dimensional, and noisy real-world time-series data are often difficult to model because the dynamics are either too complex or unknown. Traditional shallow methods, which contain a small number of non-linear operations, might not have the capacity to accurately model such complex systems.

We develop and verify three different approaches to represent and model time-series. Time-Warping SAX and Pooling SAX are two extensions of the vanilla SAX approach that is used as a symbolic representation of time series. Time-Warping SAX extracts linear temporal dependencies by building a time-delay embedding vector to construct more informative SAX words. Pooling SAX applies a non-parametric weighting scheme to extract significant variables. These are data adaptive models that achieve state-of-the-art accuacy on time-series classification problems.

We also propose the Gramian Angular Field (GAF) and Markov Transition Field (MTF) as two novel approaches to encode a time-series as an image. These representations not only demonstrate potential for visual inspection by humans, but when they are combined with deep learning approaches (Convolutional Network and Denoised Auto-encoders) they achieve excellent performance compared to other modern algorithms on classification and regression/imputation problems. GAF and MTF are non-data adaptive approaches that allow us to learn models and extract the abstract representations supported by model-based approaches.

Finally, we propose to model time-series by learning the representation directly from the raw data with model-based approaches. We will develop recurrent auto-encoders, in which the global optimum is ensured by a new Adaptive Risk-Averting/Seeking Criterion, to model the real/complex time series (dynamical systems) by learning the implicit data generating distribution over time. This model will be applied to tasks such as classification, regression/imputation, and anomaly detection.

Committee: Drs. Timothy Oates (Chair), James Lo (Math), Yun Peng and Matt Schmill

Ph.D. Proposal

Increased Autonomy with Robotics for Daily Living

Kavita Preethi Krishnaswamy

5:30pm-7:30pm, Tuesday, 9 December 2014, ITE 325B

Live Webcast: http://goo.gl/5JmjlR or http://youtu.be/qu8S6IUsCa0

Robotic technologies can provide people with disabilities invaluable tools to perform activities of daily living (ADLs). Few studies have investigated how effective and accessible the control of robotic aids is for people with severe physical disabilities with respect to their needs and current facility with technology. Though present-day robotic aids can help people with disabilities with important daily living tasks, there is still room for improvement.

What has been needed, and heretofore unavailable, is a self-directed transferring, repositioning, and personal care robotic device that is capable of increasing independence for people with physical disabilities without the assistance of caregivers. This thesis proposal will serve as the base of the research study to design and develop self-directed transferring, repositioning, and personal care robotic systems with a focus on accessible user interfaces for control that are feasible for persons with severe physical disabilities. The interface should allow local and remote control, and thus must be aware of network constraints to ensure safe and accurate control.

Committee Members: Dr. Tim Oates (Chair), Dr. Dan Ding, Dr. Tim Finin, Dr. Charles Nicholas, Dr. Yelena Yesha

PhD Dissertation Proposal

A quantum approach to the graph isomorphism
and knot classification problems

Omar Shehab

11:30am Monday, 08 December 2014, ITE 346

Simulating physics on a quantum computer can be reduced to solving mathematical problem using quantum mechanics. In this PhD dissertation proposal, I present two important mathematical problems to be investigated using quantum mechanical techniques. In the case of the first problem, the Graph Isomorphism problem, I, with another co-researcher, Kenneth M Zick, are able to present a quantum annealing algorithm with promising result. I am proposing to extend this idea so that we could study the Hamiltonian complexity of graph isomorphism. I propose to transform our previous work, the compact objective function, into a k-local Hamiltonian problem and investigate quantum adiabatic algorithms to solve it. In the case of the second problem, classification of knots, I present a mathematical framework with two restricted problems solved using quantum annealing. The restricted problems are defined for unknots embedded in two dimensional integer lattice. I propose to generalize them to higher dimension and complexity. I present how I envision to continue the rest of the study while writing the dissertation.

Committee: Drs. Samuel J. Lomonaco Jr. (chair), Alan T. Sherman, Yanhua Shih, William Gasarch

In 2012-2017, UMBC will support a total of 22 new Cybersecurity students at the BS, MS, MPS, and PhD levels in computer science and related fields. Each scholarship is for the final two years of study (three years for PhD and combined BS/MS). Each scholarship covers full tuition, fees, travel, books, and an academic year stipend of $30,000 for PhD, $25,000 for MS, and $20,000 for BS students.

Interested full-time degree students should submit an application to Dr. Alan T. Sherman, as explained on the CISA website. The same application form is used for the Scholarship For Service (SFS) and Information Assurance Scholarship Program (IASP) scholarships. Clearly state on the cover page to which program you are applying. Be sure to include official transcripts and original signed letters of reference on letterhead (preferably from tenure-track faculty who can comment on your research potential and accomplishments).

The applications must be received by the deadlines: 12noon Monday, November 17, 2014, for scholarships beginning in spring 2015 and 12noon Monday, February 2, 2015, for scholarships beginning fall 2015.

We expect to make up to two new SFS awards for spring 2015, and up to six new SFS awards for fall 2015. We do not yet know if any IASP scholarships will be possible for fall 2015.

Applicants must be US citizens capable of obtaining a secret or top-secret clearance. Each scholar must work for the federal, state, or tribal government (for pay) for one year for each year of award. Each scholar must also carry out an appropriate cybersecurity summer internship (for pay) for each year of support.

Recipients are expected to engage vigorously in cybersecurity education, research, and other cybersecurity activities while at UMBC.

For more information, contact Dr. Alan T, Sherman, Director, UMBC Center for Information Security and Assurance, .

Intern Scientist, Yahoo Labs (Job Number: 1450702)
Apply

Yahoo Labs sets the course for the future. We’re Yahoo’s incubator for bold scientific experimentation. We specialize in deep, creative thought on the company’s hardest technical problems, and hire amazing research scientists and engineers who serve as Yahoo’s most forward-looking thinkers. We stretch the limits of theory, we apply novel ideas in practice, and we experiment. We love challenging problems, we love bleeding-edge technology, and we love data.

Yahoo Labs is one of the most highly collaborative places you’ll ever see. Partnerships abound within our own teams, with all key product teams, and with the international scientific community. We’ll challenge your brain every day, and when you succeed you’ll change the lives of hundreds of millions of people, allowing them to do things they never dreamed they could do. Come join us and use your scientific background to drive Yahoo-scale innovation!

A Little About Us

Yahoo Labs is pioneering the new sciences underlying the Web. As the center of scientific excellence for Yahoo, Yahoo Labs delivers both fundamental and applied scientific leadership through published research and new technologies powering the company’s products.

Your Opportunity

We are looking for exceptional PhD student who want to work with us in our intern program for the summer of 2015. We will have openings in the US (New York City, Sunnyvale) and other locations. We seek world-class graduate students in pursuit of a PhD in Computer Science, Mathematics, Statistics, or a related area. We are particularly interested in students working on Machine Learning, algorithms, Natural Language Processing, Knowledge Representation, HCI, Multimedia, Mobile Innovations, search (systems or algorithms), collaborative filtering, auctions, mechanism design, linear algebra, Systems or analysis of large data. Ideal candidates will have finished at least 2 years of graduate work.

Your Day

• Work with scientists to perform original research
• Apply scientific thinking and techniques to improve the performance
  and effectiveness of our products
• Solve problems for our users and advertisers by analyzing mountains
  of data
• Have the opportunity to publish your work and expand the horizons
  of web science

You Must Have

• Currently working on the PhD degree, preferably in Computer
  Science, Mathematics, Statistics, or related area.
• Finished at least 2 years of graduate work
• Have some experience in Machine Learning, algorithms, Natural
  Language Processing, Knowledge Representation, HCI, Multimedia,
  Mobile Innovations, search (systems or algorithms), collaborative
  filtering, auctions, mechanism, design, linear algebra, systems, or
  analysis of large data Learning, algorithms, Natural Language
  Processing, Knowledge Representation, HCI, Multimedia, Mobile
  Innovations, search (systems or algorithms), collaborative
  filtering, auctions, mechanism, design, linear algebra, systems, or
  analysis of large data
• A CV with strong recommendations from your graduate advisor

Apply

POSITION: Verisign Labs Internships; Summer 2015

Verisign is the entity responsible for running .COM and .NET, two of DNS’ root instances, and a number of other top level domains; answering 80+ billion queries daily. Not just a leader in the domain industry, the company is also a prominent provider of DDOS protection and a suite of security services.

Verisign Labs, founded just a few years ago, is the operational research lab that generates insights from this wealth of Internet-scale routing and security data. Proprietary data sets, a computing infrastructure built for ‘big data’, and a diverse team of expert researchers are among our most prominent assets.

We are seeking highly motivated summer interns to work with our research scientists on existing and emerging topic areas in our research portfolio. Interns work closely with one or more mentors on a well-defined topic area. Past years’ intern work has resulted in internal product advancements, intellectual property (i.e., patents), and externally facing conference/journal publications. Given Verisign’s broad business capabilities and the diversity of our scientists’ expertise, the Lab seeks interns from a breadth of disciplines. Current projects include DNS privacy/security/stability, large-scale data analytics, DDoS defense, reputation and behavioral modeling, NLP over domain names, and future Internet design — and this is just a small sampling of the topics that our researchers tackle.

In addition to research work, interns participate in programming that embraces the technical, business, and social opportunities the summer provides. These include technical talks, research huddles, social events, and an intern-specific speaker series. The summer concludes with a poster session that gives interns an opportunity to present their work to Verisign’s thought and business leaders.

Verisign Labs is located in Verisign, Inc.’s global headquarters, in Reston, Virginia. The office is in Reston Town Center, an exciting cultural area with a great social atmosphere and close to plenty of outdoor activities. Reston is just about 20 miles from Washington, DC (with metro access), and is a great place to spend an exciting summer doing cutting edge research and relaxing in (and around) the nation’s capital.

Responsibilities:

• Work with technology and business leaders to identify pertinent research questions.
• Perform research tasks under the guidance of a Verisign thought leader.
• Document and present research results in appropriate forums (internal and external).

Qualifications:

• Current enrollment in a US-based PhD or MS degree program in Computer Science or a related field; PhD students are preferred.
• Availability for ~12 weeks of summer 2015 and a willingness to relocate to Reston, VA for that period. There is some flexibility with respect to start dates and duration.
• Involvement in applied/operational research; publications are a significant plus (send a complete CV).
• Experience in one or more of the following technology areas:
  • Internet infrastructure and application protocols, including TCP/IP, DNS, BGP, and HTTP.
  • Internet security and privacy, including DDoS, DNSSEC, and authentication.
  • Internet profiling and measurement, including crawling and content analysis.
  • Applied machine-learning, especially the extraction of metadata and NLP features.
  • Big data technologies (Hadoop, Hive, Pig) and statistical analysis.
  • Strong interpersonal and communications skills.

LEARN MORE here

APPLY DIRECTLY here

See Flyer here

Dissertation Defense

Virtual Circuit Provisioning in Challenged Sensor Internetworks:
with Application to the Solar System Internet

Ed Birrane

10:00am-12:00pm Monday, 11 August 2014, ITE325b

In this thesis, we present a challenged sensor internetwork (CSI) networking architecture which federates heterogeneous constituent networks behind an overlay routing mechanism abstracted from individual data link layers. The CSI is unique and required to implement expanding sensor networks.

Demand for sensing networks with increasing spatial footprints is evidenced by ongoing efforts to build geo-political border monitoring networks, intelligent highway initiatives, automated undersea surveillance, and NASA effort to construct a Solar System Internet. Existing network technologies fail to address multiple physical links, frequent disruptions, and significant signal propagation delays. The construction and maintenance of virtual circuits in an internetwork abstracted from differences in the physical, data-link, and transport layers of an internetwork represents a unique research contribution with immediate utility for a wide variety of sensing network concepts.

We describe the CSI architecture as the intersection of wireless, delay-tolerant, and heterogeneous networks and describe special characteristics of this architecture than enable useful assumptions to optimize messaging. We define an internetwork routing (INR) framework that decomposes the routing function into discrete logical steps and we provide algorithms for each of these steps. An inferred Contact Graph Routing (iCGR) algorithm populates logical graphs from local nodes. A Contact Graph Routing with Extension Blocks (CGR-EB) algorithm provides a hybrid source-path algorithm for synchronizing link state along network paths. A Predictive Capacity Consumption (PCC) algorithm exploits CGR-EB data to build a congestion model. Payload Aggregation and Fragmentation (PAF) and Traffic-Shaping Contacts (TSC) algorithms condition data and place limits on the amount of internetwork traffic carried over local networks.

From simulation, iCGR performs within ~15% of a perfect-knowledge system. CGR-EB has a speedup over standard approaches by 300% in stable topologies, by 3000% in unstable topologies, and by 11000% in unstable topologies with non-monotonic cost functions. PCC delivers 97% more data in congested networks over table-based approaches and 37% more data than the INR framework without the congestion model. PAF/TSC reduces message count by 43% while increasing goodput by 63%.

Together, these algorithms build and monitor virtual circuits in the CSI architecture. Portions of this work are in consideration for deployment in NASA networks.

Committee: Drs. Alan Sherman (Co-Chair, UMBC), Mohammed Younis (Co-Chair, UMBC), Dhananjay Phatak (UMBC), Vinton Cerf (Google), Keith Scott (MITRE), Hans Kruse (OU)