Ph.D. Dissertation Defense

Identifying and Ordering Scalar Adjectives using Lexical Substitution

Bryan Wilkinson

1:00pm Friday, 18 August 2017, ITE 325b, UMBC

Lexical semantics provides many important resources in natural language processing, despite the recent preferences for distributional methods. In this dissertation we investigate an under-represented lexical relationship, that of scalarity. We define sclarity as it relates to adjectives and introduce novel methods to identify words belonging to a particular scale and to order those words once they are found. This information has important uses in both traditional linguistics as well as natural language processing. We focus on solving both these problems using lexical substitution, a technique that allows us to determine the best substitute word for a given word in a sentence. We also produce two new datasets: a gold standard of scalar adjectives for use in the development and evaluation of methods like the ones introduces here, and a test set of indirect question-answer pairs, one possible application of scalar adjectives.

Committee: Drs. Tim Oates, CSEE (Chair), Charles Nicholas, Tim Finin, Shimei Pan (IS) and Mona Diab (GWU CS)

Computer Agents that Interact Proficiently with People

Prof. Sarit Kraus
Deptartment of Computer Science, Bar-Ilan University
Ramat-Gan, 52900 Israel

12:00-1:00pm Friday, 4 August 2017, ITE ITE 217B, UMBC

Automated agents that interact proficiently with people can be useful in supporting, training or replacing people in complex tasks. The inclusion of people presents novel problems for the design of automated agents strategies. People do not necessarily adhere to the optimal, monolithic strategies that can be derived analytically. Their behavior is affected by a multitude of social and psychological factors. In this talk I will show how combining machine learning techniques for human modeling, human behavioral models, formal decision-making and game theory approaches enables agents to interact well with people. Applications include intelligent agents that help drivers reduce energy consumption, agents that support rehabilitation, employer-employee negotiation and agents that support a human operator in managing a team of low-cost mobile robots in search and rescue task

Sarit Kraus (Ph.D. Computer Science, Hebrew University, 1989) is a Professor and is the Department Chair of Computer Science at Bar-Ilan University. Her research is focused on intelligent agents and multi-agent systems (including people and robots). In particular, she studies the development of intelligent agents that can interact proficiently with people. She studies both cooperative and conflicting scenarios. She considers modeling human behavior and predicting their decisions necessary for facing these challenges as well as the development of formal models for the agent’s decision making. She has also contributed to the research on agent optimization, homeland security, adversarial patrolling, social networks and nonmonotonic reasoning.

For her pioneer work she received many prestigious awards. She was awarded the IJCAI Computers and Thought Award, the ACM SIGART Agents Research award, the EMET prize and was twice the winner of the IFAAMAS influential paper award. She is an ACM, AAAI and ECCAI fellow and a recipient of the advanced ERC grant. She also received a special commendation from the city of Los Angeles, together with Prof. Tambe, Prof. Ordonez and their USC students, for the creation of the ARMOR security scheduling system. She has published over 350 papers in leading journals and major conferences. She is the author of the book “Strategic Negotiation in Multiagent Environments” (2001) and a co-author of the books “Heterogeneous Active Agents” (2000) and “Principles of Automated Negotiation” (2014). Kraus is a senior associate editor of the Annals of Mathematics and Artificial Intelligence Journal and an associate editor of the Journal of Autonomous Agents and Multi-Agent Systems and of JAIR. She is a member of the board of directors of the International Foundation for Multi-agent Systems (IFAAMAS).

Baltimore Sun highlights UMBC programs that prepare students for high-demand careers

The latest special section on education in The Baltimore Sun highlights several UMBC programs that prepare students to succeed in careers in rapidly growing and already high-demand industries. The Sun highlights how the flexibility of these programs makes them particularly accessible and valuable to students, allowing students to tailor their pathway to match specific areas of interest.

Marc Olano, associate professor of computer science and electrical engineering and director of the game development track in computer science, described UMBC’s multiple pathways for learning game development. “We created two programs, the game development track in computer science and the animation and interactive media concentration in visual arts,” he explained. “In both cases, the students get a degree in the primary discipline, a B.S. in computer science or a B.A. or B.F.A. in visual arts,” he noted. The “track” system helps students focus their electives specifically on skills “that are used in the games industry.”

Jacqueline Wojcik ‘17, visual arts, who completed a concentration in animation and interactive media and a minor in computer science, shared her experience as a member of UMBC’s Game Developers Club. Collaborating with other students in the club helped her apply skills she learned in the classroom to other creative opportunities. For the coming year, Wojcik has received a Fulbright research grant to complete an innovative project in Oslo, Norway. “I will create digital models from two Viking Age ship burials and then place in interactive, game-like environments so that people can see how the artifacts would have used in the lives of their owners,” she told the Sun. “The project will explore the intersection of games, learning and archeological visualization.”

The education section also highlighted UMBC programs in health information technology and cybersecurity.

The health IT program prepares professionals with backgrounds in computer science, information systems, and health care for growing opportunities in work to prevent medical errors, improve care delivery, and address other major health care challenges through technology. “Not only does it provide students with technical information technology skills, but it also provides a pragmatic understanding of the health care system, which has its own terminology,” explains Krystl Haerian ‘99, biological science, an instructor in the program.

Laura Humber ‘16, health administration and policy, a current student in the health IT master’s program, says that she didn’t think she would end up pursuing an advanced degree in such a technical-sounding field, but she came to realize it could help her take the next steps in a research career focused on addressing opioid addiction. “People think when you say, ‘IT’ that you’re dealing with computers, but it’s making computers work for you and get you the information you need,” Humber explained.

Rick Forno, assistant director of UMBC’s Center for Cybersecurity and director of the graduate cybersecurity program, described in the Sun how UMBC’s cybersecurity graduate programs are also designed work across fields and to connect students with skills that will help them advance their careers. He noted that at UMBC, studying cybersecurity can include courses in economics, public policy, and biotechnology. “Pick a field or major, and cybersecurity applies to it,” he said. “It really is interdisciplinary.”

Students are encouraged to tailor the program to their interests, and gain experience through research and internship opportunities, so they can explore the full range of career opportunities available to them.“You can go to grad school and get a degree and get a job,” Forno reflected. “But we want you to be a professional. You can do more than just your degree. There’s depth to what we offer.”

Read the full education section in The Baltimore Sun.

This post was adapted from a UMBC News artcle written by Megan Hanks. Banner image: The game developers club display in House of Grit at UMBC’s 50th Anniversary celebration. Photo by freelance photographer.

UMBC Ph.D. candidate Kavita Krishnaswamy receives
Google and Microsoft awards for robotics research

Kavita Krishnaswamy ’07, computer science and mathematics, Ph.D. ’18, computer science, has been named both a 2017 Microsoft Fellow and recipient of the Google Lime Scholarship. These prestigious honors recognize emerging scholars in computing who are dedicated to increasing diversity in the field, and Krishnawamy’s awards will support her Ph.D. research on “Smart Algorithms via Knowledge Management of Safe Physical Human-Robotic Care.”

“I am very humbled, honored, and grateful to Google Lime and Microsoft for providing me with this enriching experience and lifetime opportunity to serve society by advancing the field of human-robot interaction,” Krishnaswamy says.

The Google Lime Scholarship seeks to promote greater access to knowledge for people with visible and invisible disabilities. It was established through a partnership between Google and Lime Connect, a nonprofit focused on breaking stereotypes about disability and encouraging companies to recognize the importance and value of employing people with disabilities.

The program encourages students with disabilities to pursue their passions in computing and technology, and to become leaders in those fields. As part of the fellowship program, Krishnaswamy will receive scholarship funding and will participate in the 2017 Google Scholars’ Retreat.

Through the Microsoft Fellowship, Krishnaswamy will receive funding to support her Ph.D. research and will participate in the Microsoft Research workshop held in fall 2017. Her research currently focuses on building a teleoperated mobile robotic prototype, in addition to creating an accessible robotic interface, that seniors and people with disabilities will be able to control by repositioning their arms and legs. Tim Oates, professor of computer science and electrical engineering, is Krishnaswamy’s Ph.D. advisor.

“Our goal is to explore the intersection between providing physical care and robotics, and how it is possible to translate safe patient handling and mobility guidelines into smart human-robotic interaction algorithms,” Krishnaswamy explains. “As assistive robotics become more mainstream, these best practices can improve safety in direct physical care in the process of repositioning the human body with a mobile robotic arm.”

Krishnaswamy is excited about the possible new directions her research can now explore, thanks to support from these awards. “The resources will provide me with a solid and steady foundation to cultivate new technical expertise and professional skills to successfully continue my dissertation research in robotics, and to broaden my knowledge in the field,” she says.

Krishnaswamy has been recognized internationally as an emerging leader in robotics and accessibility design throughout her graduate studies. She is a former Ford Foundation Predoctoral Fellow, and a National Science Foundation Graduate Research Fellow. In 2015, she was named to Robohub’s “25 Women in Robotics You Need to Know About” list.

This post was adapted from a UMBC News article written by Megan Hanks. Image by Kavita Krishnaswamy.

Professor Cynthia Matuszek has received a research award from the National Science Foundation to improve human-robot interactions by enabling them to understand the world from natural language in order to take instructions and learn about their environment naturally and intuitively. The two-year award, Joint Models of Language and Context for Robotic Language Acquisition, will support Dr. Matuszsek’s Interactive Robotics and Language Lab, which focuses on how robots can flexibly learn from interactions with people and environments.

As robots become smaller, less expensive, and more capable, they are able to perform an increasing variety of tasks, leading to revolutionary improvements in domains such as automobile safety and manufacturing. However, their inflexibility makes them hard to deploy in human-centric environments, such as homes and schools, where their tasks and environments are constantly changing. Meanwhile, learning to understand language about the physical world is a growing research area in both robotics and natural language processing. The core problem her research addresses is how the meanings of words are grounded in the noisy, perceptual world in which a robot operates.

The ability for robots to follow spoken or written directions reduces the adoption barrier for robots in domains such as assistive technology, education, and caretaking, where interactions with non-specialists are crucial. Such robots have the potential to ultimately improve autonomy and independence for populations such as aging-in-place elders; for example, a manipulator arm that can learn from a user’s explanation how to handle food or open novel containers would directly affect the independence of persons with dexterity concerns such as advanced arthritis.

Matuszek’s research will investigate how linguistic and perceptual models can be expanded during interaction, allowing robots to understand novel language about unanticipated domains. In particular, the focus is on developing new learning approaches that correctly induce joint models of language and perception, building data-driven language models that add new semantic representations over time. The work will combines semantic parser learning, which provides a distribution over possible interpretations of language, with perceptual representations of the underlying world. New concepts will be added on the fly as new words and new perceptual data are encountered, and a semantically meaningful model can be trained by maximizing the expected likelihood of language and visual components. This integrated approach allows for effective model updates with no explicit labeling of words or percepts. This approach will be combined with experiments on improving learning efficiency by incorporating active learning, leveraging a robot’s ability to ask questions about objects in the world.

Dissertation Defense

The Lightweight Virtual File System

Navid Golpayegani

10:00-12:00 Thursday, 20 July 2017, ITE 325, UMBC

A data center today is responsible for safely managing big data volumes and balancing the complex needs between data producers and consumers. This balance often involves reconciling the needs of easy access and rapid retrieval in ways desired by the consumers with the needs of long term availability, reliability, and expandability of data producers. The long term continuous support of data storage adds another layer of complexity for the file system. As storage architecture and big data volumes evolve, existing file system’s primary focus is performance while less attention is payed to addressing the problems of the above long term servicing needs of their clients.

I have developed the Lightweight Virtual File System (LVFS) to address these problems through the unique conceptual approach of separating the most common tasks involved in a file system; namely storing data, locating data, and organizing data. Standard file systems are developed as single monolithic systems performing all three tasks. LVFS replaces these tasks with an architecture which enables the dynamic combination of different algorithms for each of those tasks. Using this approach, LVFS is capable of constructing a storage system, which allows for ready availability, reliability, expandability, and long term support while, simultaneously, assuring the performance of a stable system customizable to meet the needs of data consumers.

After successful development and testing to allow for merging decades old storage architecture with new and incompatible ones, such as HGST Active Archive System, NASA Goddard Space Flight Center’s Terrestrial Information Systems Laboratory adopted LVFS for their production environment to create a single, integrated storage system without any software modifications. UMBC’s Center for Hybrid Multicore Productivity Research deployed an instance on the IBM iDataPlex ‘BlueWave’ cluster to utilize Seagate’s Active Drive systems as a storage and on-disk compute platform. With LVFS we show we were able to perform MapReduce computation directly on the drive with comparable performance to Hadoop running on BlueWave. It also shows a significant reduction in data leaving the active drive during computation thereby significantly increasing throughput.

Committee Members: Dr.s Milton Halem (Advisor), Yelena Yesha, John Dorband, Charles Nicholas, Curt Tilmes

Dissertation Defense

Deep Representation of Lyrical Style and Semantics for Music Recommendation

Abhay L. Kashyap

11:00-1:00 Thursday, 20 July 2017, ITE 346

In the age of music streaming, the need for effective recommendations is important for music discovery and a personalized user experience. Collaborative filtering based recommenders suffer from popularity bias and cold-start which is commonly mitigated by content features. For music, research in content based methods have mainly been focused in the acoustic domain while lyrical content has received little attention. Lyrics contain information about a song’s topic and sentiment that cannot be easily extracted from the audio. This is especially important for lyrics-centric genres like Rap, which was the most streamed genre in 2016. The goal of this dissertation is to explore and evaluate different lyrical content features that could be useful for content, context and emotion based models for music recommendation systems.

With Rap as the primary use case, this dissertation focuses on featurizing two main aspects of lyrics; its artistic style of composition and its semantic content. For lyrical style, a suite of high level rhyme density features are extracted in addition to literary features like the use of figurative language, profanity and vocabulary strength. In contrast to these engineered features, Convolutional Neural Networks (CNN) are used to automatically learn rhyme patterns and other relevant features. For semantics, lyrics are represented using both traditional IR techniques and the more recent neural embedding methods.

These lyrical features are evaluated for artist identification and compared with artist and song similarity measures from a real-world collaborative filtering based recommendation system from Last.fm. It is shown that both rhyme and literary features serve as strong indicators to characterize artists with feature learning methods like CNNs achieving comparable results. For artist and song similarity, a strong relationship was observed between these features and the way users consume music while neural embedding methods significantly outperformed LSA. Finally, this work is accompanied by a web-application, Rapalytics.com, that is dedicated to visualizing all these lyrical features and has been featured on a number of media outlets, most notably, Vox, attn: and Metro.

Committee: Drs. Tim Finin (chair), Anupam Joshi, Tim Oates, Cynthia Matuszek and Pranam Kolari (Walmart Labs)

Analysis of Irregular Event Sequences using Deep Learning, Reinforcement Learning, and Visualization

Filip Dabek

11:00-1:00 Thursday 13 July 2017, ITE 346, UMBC

History is nothing but a catalogued series of events organized into data. Amazon, the largest online retailer in the world, processes over 2,000 orders per minute. Orders come from customers on a recurring basis through subscriptions or as one-off spontaneous purchases, resulting in each customer exhibiting their own behavioral pattern when it comes to the way in which they place orders throughout the year. For a company such as Amazon, that generates over $130 billion of revenue each year, understanding and uncovering the hidden patterns and trends within this data is paramount in improving the efficiency of their infrastructure ranging from the management of the inventory within their warehouses, distribution of their labor force, and preparation of their online systems for the load of users. With the ever increasingly availability of big data, problems such as these are no longer limited to large corporations but are experienced across a wide range of domains and faced by analysts and researchers each and every day.

While many event analysis and time series tools have been developed for the purpose of analyzing such datasets, most approaches tend to target clean and evenly spaced data. When faced with noisy or irregular data, it has been recommended to undergo a pre-processing step of converting and transforming the data into being regular. This transformation technique arguably interferes on a fundamental level as to how the data is represented, and may irrevocably bias the way in which results are obtained. Therefore, operating on raw data, in its noisy natural form, is necessary to ensure that the insights gathered through analysis are accurate and valid.

In this dissertation novel approaches are presented for analyzing irregular event sequences using a variety of techniques ranging from deep learning, reinforcement learning, and visualization. We show how common tasks in event analysis can be performed directly on an irregular event dataset without requiring a transformation that alters the natural representation of the process that the data was captured from. The three tasks that we showcase include: (i) summarization of large event datasets, (ii) modeling the processes that create events, and (iii) predicting future events that will occur.

Committee: Drs. Tim Oates (Chair), Jesus Caban, Penny Rheingans, Jian Chen, Tim Finin

Name: Alex Hart

Educational Background: Bachelor’s degree in Accounting from the University of Maryland, College Park

Hometown: Baltimore, MD (Go O’s and Ravens!)

Current role: As an Accountant I, Alex provides business services support to the CSEE department in the areas of contracts and grants/projects, which includes account monitoring, financial reporting, projections, reconciliations, etc. She also provides backup support for payroll, and she is the property custodian of inventory for CSEE.

Favorite thing about UMBC: “Without a doubt, my favorite thing about UMBC is the people here. I have met a lot of different people who have provided me with a wealth of knowledge since I started working here just a year ago. Everyone has been very inclusive and helpful!”

Students should ask me about: “Students can ask me anything, but maybe about the college experience, since I’m still a recent graduate.”

Alex is originally from Baltimore, MD. She joined CSEE’s Department in February of 2016. She attended UMBC for her first two years of college, then transferred to the University of Maryland College Park’s Robert H. Smith School of Business. She has a BS in Accounting from UMCP.

When not working, Alex loves cheering on the Terps in football and basketball. She also enjoys traveling to new places, cooking, practicing yoga, and reading.

Every day, people interact with large amounts of text online, including legal documents they might quickly skim and sign without full, careful review. In an article recently published in The Conversation, Karuna Joshi, research associate professor of computer science and electrical engineering, and Tim Finin, professor of computer science and electrical engineering, explain how artificial intelligence (AI) is helping to summarize lengthy and complex legalese so people can more easily understand terms of service and similar agreements before they click “accept” to access a new app or online service.

The legal documents that Joshi and Finin are working to summarize—terms of service, privacy policies, and user agreements—often accompany new online services, contests, apps, and subscriptions. “As computer science researchers, we are working on ways artificial intelligence algorithms could digest these massive texts and extract their meaning, presenting it in terms regular people can understand,” they explain.

Through their research, Joshi and Finin ask computers to break down the terms and conditions that regular users “agree” to or “accept.” To process the text, Joshi and Finin employ a range of AI technologies, including machine learning, knowledge representation, speech recognition, and human language comprehension.

Joshi and Finin have found that in many of the privacy policies people are prompted to review and accept online, there are sections that do not actually apply to the consumer or service provider. These sections of the agreements might, for example, “include rules for third parties…that people might not even know are involved in data storage or retrieval,” they note.

After examining these documents, the software Joshi and Finin have developed pinpoints specific items that people should be aware of when they are granting their consent or agreement—what they describe as “key information specifying the legal rights, obligations and prohibitions identified in the document.” In other words, the software takes in all that complex legal language, and then then presents just the most essential information in “clear, direct, human-readable statements,” making it much more feasible for users to understand what they are consenting to before they click “agree.”

Read “Teaching machines to understand — and summarize — text” in The Conversation to learn more about Joshi and Finin’s approach to making online legal documents more accessible through AI.

Adapted from a UMBC News article by Megan Hanks Banner image: Karuna Joshi. Photo by Marlayna Demond ’11 for UMBC.