Lyrics Augmented Multi-modal
Music Recommendation

Abhay Kashyap

1:00pm Friday 30 October, ITE 325b

In an increasingly mobile and connected world, digital music consumption has rapidly increased. More recently, faster and cheaper mobile bandwidth has given the average mobile user the potential to access large troves of music through streaming services like Spotify and Google Music that boast catalogs with tens of millions of songs. At this scale, effective music recommendation is critical for music discovery and personalized user experience.

Recommenders that rely on collaborative information suffer from two major problems: the long tail problem, which is induced by popularity bias, and the cold start problem caused by new items with no data. In such cases, they fall back on content to compute similarity. For music, content based features can be divided into acoustic and textual domains. Acoustic features are extracted from the audio signal while textual features come from song metadata, lyrical content, collaborative tags and associated web text.

Research in content based music similarity has largely been focused in the acoustic domain while text based features have been limited to metadata, tags and shallow methods for web text and lyrics. Song lyrics house information about the sentiment and topic of a song that cannot be easily extracted from the audio. Past work has shown that even shallow lyrical features improved audio-only features and in some tasks like mood classification, outperformed audio-only features. In addition, lyrics are also easily available which make them a valuable resource and warrant a deeper analysis.

The goal of this research is to fill the lyrical gap in existing music recommender systems. The first step is to build algorithms to extract and represent the meaning and emotion contained in the song’s lyrics. The next step is to effectively combine lyrical features with acoustic and collaborative information to build a multi-modal recommendation engine.

For this work, the genre is restricted to Rap because it is a lyrics-centric genre and techniques built for Rap can be generalized to other genres. It was also the highest streamed genre in 2014, accounting for 28.5% of all music streamed. Rap lyrics are scraped from dedicated lyrics websites like ohhla.com and genius.com while the semantic knowledge base comprising artists, albums and song metadata come from the MusicBrainz project. Acoustic features are directly used from EchoNest while collaborative information like tags, plays, co-plays etc. come from Last.fm.

Preliminary work involved extraction of compositional style features like rhyme patterns and density, vocabulary size, simile and profanity usage from over 10,000 songs by over 150 artists. These features are available for users to browse and explore through interactive visualizations on Rapalytics.com. Song semantics were represented using off-the-shelf neural language based vector models (doc2vec). Future work will involve building novel language models for lyrics and latent representations for attributes that is driven by collaborative information for multi-modal recommendation.

Committee: Drs. Tim Finin (Chair), Anupam Joshi, Pranam Kolari (WalmartLabs), Cynthia Matuszek and Tim Oates

The UMBC CSEE Seminar Series Presents

Graphical-model-based machine learning for neuroimaging data

Professor Rong Chen
University of Maryland School of Medicine

12noon-1pm Friday, 30 October 2015, ITE 102, UMBC

Two important problem in neuroimaging data mining is high-dimensionality and temporal network modeling. Analyzing high-dimensional neuroimaging data is a very challenging problem. We developed an algorithms called Graphical-Model-based Multivariate Analysis (GAMMA) to model complex interactions among brain regions and a clinical variable. GAMMA has embedded dimension reduction and regularization mechanism. GAMMA has been used in distinguishing patients with mild cognitive impairment and normal elderly.

Identifying spatial-temporal interactions among brain regions from longitudinal structural magnetic-resonance images presents one of the major challenges in computational neuroanatomy. We developed a dynamic Bayesian network based method called structural dynamic network analysis (SDNA) to solve this problem. SDNA enables the detection of spatial-temporal interactions among brain regions, leading to dynamic network analysis. SDNA has been used to model trajectory changes in patients with Alzheimer’s disease.

Dr. Rong Chen is an Assistant Professor at in the department of Radiology the University of Maryland School of Medicine. He completed his Ph.D. in Electrical and Computer Engineering at Washington State University in 2003, and his MTR in Translational Medicine at the University of Pennsylvania in 2012. He published 45 peer-reviewed research articles in the areas of neuroimaging and data mining. His research interests include computational neuroscience, data mining, medical image analysis, and translational medicine.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

About the CSEE Seminar Series: The UMBC Department of Computer Science and Electrical Engineering presents technical talks on current significant research projects of broad interest to the Department and the research community. Each talk is free and open to the public. We welcome your feedback and suggestions for future talks.

WEBINAR: Chinese Cyber Power

6:00-7:00pm Monday, 26 Oct 26 2015
Register online

Dr. Terry Thompson, of the UMBC Cybersecurity Graduate Program faculty, will present on the political, economic, military, and foreign policy dimensions of China’s cyber strategy and operations. He will explore:

• How does cyber fit into China’s military strategy?
• Why is there so much focus on offensive cyber operations?
• What is the Chinese view of the U.S. response (or lack of responses) to their cyber attacks on U.S. government and industry?
• Given the large amount of U.S. debt held by China, what is the rationale for cyber attacks that can damage the U.S. economy?
• Who is behind the strategy and operational planning and execution of China’s cyber attacks on the U.S.?

This webinar is a preview of Dr. Thompson’s Spring 2016 course CYBR 691 Special Topics in Cybersecurity: “Chinese Cyber Power: Perspectives and Implications.” Please Note: Due to time constraints, not all topics may be covered during the webinar.

Register online

CHMPR Distinguished Lecture Series

Programming and Tuning a Quantum Annealing Computer to Solve Real-World Applications

Dr. Alejandro Perdomo-Ortiz

Quantum Artificial Intelligence Laboratory
NASA Ames Research Center

2:00pm Monday 26 October 2015, ITE 325b

Since September 2013 and through a partnership with Google and USRA, NASA Ames Research Center has been working with a quantum device that has the promise of harnessing quantum-mechanical effects to speed up the solution of optimization problems. Solving real-world applications with quantum algorithms requires overcoming several challenges, ranging from translating the computational problem at hand to the quantum-machine language, to tuning several other parameters of the quantum algorithm that have a significant impact on performance of the device. In this talk, we discuss these challenges, strategies developed to enhance performance, and also a more efficient implementation of several applications. Although we will focus on applications of interest to NASA’s Quantum Artificial Intelligence Laboratory, the methods and concepts presented here apply to a broader family of hard discrete optimization problems that might also be present in many machine-learning algorithms.

Alejandro Perdomo-Ortiz is a Research Scientist at NASA Ames Research Center, Quantum Artificial Intelligence Laboratory, where he works in the design of quantum algorithms to solve hard optimization problems. Alejandro received a Ph.D. in Chemical Physics from Harvard University. He is a three-time winner of Harvard’s Certificate of Excellence in Teaching and a recipient of the Dudley R. Herschbach Teaching Award. He is originally from Cali, Colombia where he performed undergraduate studies in Chemistry at Universidad del Valle. Within the NASA team, he is interested in understanding the scalability and performance of quantum annealing algorithms and their realistic experimental implementations for broad applications in space exploration research.

Host: Prof. Milton Halem, 

 The UMBC CSEE Seminar Series Presents

Online Learning for Cognitive Radios,
Power Grids, and Brain Imaging

Dr. Seung-Jun Kim
Department of CSEE, UMBC

1-2pm, Friday, 23 October 2015, ITE 325b

With the advent of big data era with pervasive sensors and powerful computational intelligence techniques, application of data-driven techniques to various domains is becoming quite popular. In this talk, some of our recent research activities in the signal processing and smart systems lab (SPSS) will be sampled. In particular, it will be highlighted how the online learning techniques can benefit different applications in the wireless communication, power systems, and medical imaging areas.

Seung-Jun Kim received his B.S. and M.S. degrees from Seoul National University in Seoul, Korea, and his Ph.D. from the University of California at Santa Barbara in 2005, all in electrical engineering. From 2005 to 2008, he worked for NEC Laboratories America in Princeton, New Jersey. He was with the University of Minnesota during 2008-2014, where his final title was Research Associate Professor. In August 2014, he joined the CSEE department at UMBC. Dr. Kim’s research interests include statistical signal processing, optimization, and machine learning, with applications to wireless communication and networking, future power systems, and big data analytics.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

About the CSEE Seminar Series: The UMBC Department of Computer Science and Electrical Engineering presents technical talks on current significant research projects of broad interest to the Department and the research community. Each talk is free and open to the public. We welcome your feedback and suggestions for future talks. Upcoming talks include the following.

Signature Track (Fridays, 12noon-1pm, in ITE 102):

• Oct. 30, Rong Chen, SOM Faculty, computational neuroscience
• Nov.13, John Kloetzli (Firaxis), computer graphics
• Weekly Track (Thursday 12noon-1pm, or Friday 1-2pm, in ITE 325):
• Nov. 20 Hamed Pirsiavash (UMBC), computer vision
• Nov. 6 Nilanjan Banerjee (UMBC), Internet of Things
• Dec. 4 Ting Zhu (UMBC), energy system and big data

Other UMBC CSEE Seminar Series: The UMBC Cyber Defense Lab (CDL) meets biweekly Fridays 11:15am-12:30pm in ITE 231, for research talks about cybersecurity.

The UMBC Cyber Defense Lab presents

Are your personal data at risk?
App analytics to the rescue

Prajit Kumar Das
Ebiquity, CSEE Department, UMBC

11:15am-12:30pm, Friday, 23 October 2015, ITE 231

According to the prominent virus and malware tool Virustotal, the Google Play Store has a few thousand apps from major malware families. Given such a revelation, access control systems for mobile data management have reached a state of critical importance. We propose developing a system that will help us detect pathways along which user data are being stolen from their mobile devices. We use a multi-layered approach including app meta data analysis, understanding code patterns, and detecting and eventually controlling dynamic data flow when such an app is installed on a mobile device. In this presentation we focus on the first part of our work and discuss the merits and flaws of our unsupervised learning mechanism to detect possible malicious behavior from apps in the Google Play Store.

Prajit Das is a PhD student in computer science at UMBC.

Host: Alan T. Sherman,

The UMBC CSEE Seminar Series Presents

Enhanced IP and OpenFlow Switching to
Provide Zero Touch Traffic Engineering

Dr. William Chimiak
Laboratory for Telecommunications Science (LTS)

12noon-1pm, Friday, 16 October, 2015, ITE 102

I propose a method using OpenFlow and Enhanced IP (64 bit IPv4) to provide an end-to-end method of creating traffic engineered flow paths for big data. With an SSL registered Northbound Application, a user with proper credentials requests an big data transfer. This is sent to a port with a hybrid Enhanced IP NAT. The Enhanced IP portion of the NAT is stateless making communication-set up faster, but allows the normal Carrier-grade NAT function, if necessary. With this system, there will be a mechanism to allow end-to-end awareness of the flow type to allow for an end-to-end traffic engineered path.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

About the CSEE Seminar Series: The UMBC Department of Computer Science and Electrical Engineering presents technical talks on current significant research projects of broad interest to the Department and the research community. Each talk is free and open to the public. We welcome your feedback and suggestions for future talks. Upcoming talks include the following.

Signature Track (Fridays, 12noon-1pm, in ITE 102):

• Oct. 30, Rong Chen, SOM Faculty, computational neuroscience
• Nov.13, John Kloetzli (Firaxis), computer graphics

Weekly Track (Thursday 12noon-1pm, or Friday 1-2pm, in ITE 325):

• Nov. 20 Hamed Pirsiavash (UMBC), computer vision
• Nov. 6 Nilanjan Banerjee (UMBC), Internet of Things
• Dec. 4 Ting Zhu (UMBC), energy system and big data

Other UMBC CSEE Seminar Series: The UMBC Cyber Defense Lab (CDL) meets biweekly Fridays 11:15am-12:30pm in ITE 231, for research talks about cybersecurity. Next talk is 10-23.

The UMBC CSEE Seminar Series Presents

Grounded Language Acquisition: A Physical Agent Approach

Dr. Cynthia Matuszek

Interactive Robotics and Language Lab
Computer Science and Electrical Engineering, UMBC

12:00-1:00pm Friday, 9 Oct. 2015, ITE 325b

A critical component of understanding human language is the ability to map words and ideas in that language to aspects of the external world. This mapping, called the symbol grounding problem, has been studied since the early days of artificial intelligence; however, advances in language processing, sensory, and motor systems have only recently made it possible to directly interact with tangibly grounded concepts. In this talk, I describe how we combine robotics and natural language processing to acquire and use physically grounded language specifically, how robots can learn to follow instructions, understand descriptions of objects, and build models of language and the physical world from interactions with users. I will describe our work on building a learning system that can ground English commands and descriptions from examples, making it possible for robots to learn from untrained end-users in an intuitive, natural way, and describe applications of our work in following directions and learning about objects. Finally, I will discuss how robots with these learning capabilities address a number of near-term challenges.

Cynthia Matuszek is an Assistant Professor at the University of Maryland, Baltimore County’s Computer Science and Electrical Engineering department. She completed her Ph.D. at the University of Washington in 2014, where she was a member of both the Robotics and State Estimation lab and the Language, Interaction, and Learning group. She is published in the areas of artificial intelligence, robotics, ubiquitous computing, and human-robot interaction. Her research interests include human-robot interaction, natural language processing, and machine learning.

Hosts: Professors Fow-Sen Choa () and Alan T. Sherman ()

· directions and more information ·

The UMBC Cyber Defense Lab presents

Hack, Play, Win: Lessons Learned Running
The Maryland Cyber Challenge

Richard Forno, UMBC

11:15am-12:30pm, Friday, 9 Oct 2015, ITE 231

An oft-cited and prominent concern facing the Internet security community is the need to identify and hire qualified cybersecurity practitioners able to fill critical technical, analytical, and managerial positions within the global technology workforce. A 2014 report from the Education Advisory Board discusses the “exploding” demand for qualified cybersecurity practitioners, noting that cybersecurity jobs grew by 73% between 2007-2012 compared to 6% in all other industry sectors. Similarly, Burning Glass, a national employment research firm, notes that there are nearly 23,000 available cybersecurity positions in the Washington, DC metropolitan area. Nowhere is this need more evident, or discussed more frequently, than in Maryland, a region some dub the ‘epicenter of cybersecurity’ education, research, and industry.

In response to this concern, events in the cybersecurity discipline, known as cyber competitions” or “cyber challenges” seek to motivate and encourage high school and college students toward careers in cybersecurity by developing their technical and teamwork skills while also allowing more experienced cybersecurity professionals an opportunity to practice their expertise in a challenging venue for professional recognition. The popularity and number of these events as a form of intellectual competition at industry security conferences like the DEFCON CTF or Department of Defense DC3 Digital Forensics Challenge and those within educational communities such as the National Cyber League (NCL), CyberPatriot, or the Collegiate CyberDefense Competition (CCDC) are but a few examples of prominent cyber challenges drawing worldwide participation. Other competitions, both large and small, are under continual development, as is a National Science Foundation-backed effort to create a national federation to support and standardize the rules, activities, and conduct of cyber competitions.

Given the popularity of these events, and the ongoing global desire to launch new ones, this talk will draw upon the experiences of organizing and coordinating the Maryland Cyber Challenge (MDC3) from 2011-2014 in offering advice to current and future cyber competition planners. What lessons from current competitions can help future competition organizers run successful challenges of their own? And are such events enough to prepare the next generation of cybersecurity professional? While no event will ever run perfectly, organizers must always strive to “get it right” – or as close to “right” as possible!

(This talk previews a paper accepted for publication in the December 2015 USENIX ;login;)

Dr. Richard Forno directs the University of Maryland, Baltimore County’s Graduate Cybersecurity Program, serves as the Assistant Director of UMBC’s Center for Cybersecurity, and is a Junior Affiliate Scholar at the Stanford Law School’s Center for Internet and Society (CIS). His twenty-year career spans the government, military, and private sector, including helping build a formal cybersecurity program for the US House of Representatives, serving as the first Chief Security Officer for the InterNIC, and co-founding the Maryland Cyber Challenge. Richard was also one of the early researchers on the subject of “information warfare” and he remains a longtime commentator on the influence of Internet technology upon society.

Host: Alan T. Sherman,

Programming Robotic Agents: A Multi-tasking Teleo-Reactive Approach

Keith Clark, Imperial College London
University of Queensland, University New South Wales
joint work with Peter Robinson, University of Queensland

2:00pm Friday, 2 October 2015, ITE325b

We present a multi-threaded/multi-tasking message communicating robotic agent architecture in which the concurrently executing tasks are programmed in TeleoR, a major extension of Nilsson’s Teleo-Reactive Procedures (TR) guard ~> action rule language for robotic agents.

The rule guards query rapidly changing percept facts, and more slowly changing told and remembered facts, using fixed facts, relation and function rules (the agent’s knowledge) in the agent’s deductive BeliefStore. Its operational semantics makes the languages well suited to robot/robot or human/robot co-operative tasks.

TeleoR extends TR in:

• being typed and higher order,
• having a typed higher order LP/FP language, QuLog, for encoding BeliefStore knowledge,
• having extra forms of rules and actions, and o having task atomic procedures to control the deadlock and starvation free sharing of several robotic resources by concurrently executing tasks.

Its use is illustrated in the video at http://bit.ly/teleor. It is being used at UNSW to write the control program for a two armed Baxter robot working in co-operation with a person concurrently engaged in several assembly tasks.

Keith Clark is Emeritus Professor of Computer Science at Imperial College London, England and a Visiting Professor at the University of Queensland and the University New South Wales. He has lectured in both mathematics and computer science.

Host: Tim Finin

Upcoming talks and directions