News featuring Chang-Tien Lu

Chang-Tien Lu working with Maryland firm on advanced event prediction system to improve metro security

Chang-Tien Lu, a professor in the Department of Computer Science and associate director of DAC

Chang-Tien Lu, a professor in the Department of Computer Science, associate director of the Discovery Analytics Center, and a faculty member in the National Science Foundation- sponsored UrbComp program  has teamed up with Schneider Electric Buildings Critical Systems, Inc. (SEBCSI)  on a project entitled “Advanced Analytics for High-Performance Metro Security Monitoring,” funded by the Washington Metropolitan Area Transit Authority (WMATA)

SEBCSI, located in Columbia, Maryland, provides electronic security solutions, including security access control and closed circuit television (CCTV) for commercial and government applications.

Lu and a team of graduate students will develop an advanced spatiotemporal event detection system of several layers to deal with data preprocessing, information extraction, threat level analysis, and visualization and extract security-related information from social media contents that can help metro police improve security on trains and at metro stations.

“With the popularity of mobile devices, more public transport riders are using Twitter to report observations of potential threats, ongoing crimes, and other metro incidents. Social media serves as a good supplemental data source because tweets can cover multiple types of security issues — the most common being fight, stealth, and robbery — that are typically posted by a witness either during an ongoing crime or by the victim immediately after a crime,” said Lu.

Another advantage of social media, said Lu, is that security issue related tweets usually include station names and very accurate timestamps.

The researchers’ new system will mine collected security issues from tweet collections and visualize a crime hotspot of metro stations on a geospatial map based on location information while simultaneously representing an auto-summarized station-wise report.

“This data can help metro police make better decisions in deploying limited resources,” Lu said.

DAC students from the Department of Computer Science who are working with Lu on this project are Ph.D. student Jianfeng He and master’s students Omer Zulfiqar, Yi-Chun Chang, and Po-Han Chen.

The project builds on Lu’s previous research in intelligent transportation systems, “Steds: Social Media based Transportation Event Detection with Text Summarization,”

“A Search and Summary Application for Traffic Events Detection Based on Twitter Data,”

“Determining Relative Airport Threats from News and Social Media,” and

 “TITAN: A Spatiotemporal Feature Learning Framework for Traffic Incident Duration.”




DAC Student Spotlight: Zhiqian “Danny” Chen

Zhiqian “Danny” Chen, DAC Ph.D. student in the Department of Computer Science

Generating new music inspired by existing music datasets is a major area of interest for Zhiqian “Danny” Chen, a Ph.D. student at the Discovery Analytics Center.

“Music, with its complex hierarchical and sequential structure and its inherent emotional and aesthetic subjectivity, is an intriguing research subject at the core of human creativity,” said Chen. “And because of rapid advances in data-driven algorithms such as deep learning, exploring computational creativity via machine learning approaches is increasingly popular.”

While this exploration has included some work on generative models for music, research that investigates the capabilities of deep learning for creative applications such as style transfer on images and video is limited, he said.

The paper, “Learning to Fuse Music Genres with Generative Adversarial Dual Learning,” aims to fill this space by exploring the idea of style fusion in music with generative adversarial dual learning. Chen presented the paper at the 2017 IEEE International Conference on Data Mining (ICDM) in New Orleans, Louisiana.

In November, Chen will attend ICDM 2018 in Singapore, where he will present “Rational Neural Networks for Approximating Jump Discontinuities of the Graph Convolution Operator,” a study on deep graph learning.

“Effective information analysis generally boils down to the geometry of the data represented by a graph,” said Chen. Typical applications include social networks, transportation networks, spread of epidemic diseases, neuronal networks, biological regulatory networks, telecommunication networks, and knowledge graphs, defined over non-Euclidean graph domains.

Chen’s research in this area focuses on modeling graphs in spectral domains for deriving representations for node level embeddings. He uses graph notions such as adjacency matrices or graph Laplacians to describe geometric structures and reveal latent patterns.

He holds a bachelor’s degree in software engineering and Japanese from Huazhong University of Science and Technology, China, and a master’s degree in software engineering from Peking University, China.

Advised by Chang-Tien Lu, Chen is projected to graduate with a Ph.D. in computer science in Fall 2019.

DAC Student Spotlight: Kaiqun Fu

Kaiqun Fu, DAC Ph.D. student in computer science

In his research, Kaiqun Fu uses spatial data mining, urban computing, and machine learning to infer crime rates/types from street view images, roadway networks, and criminal records. Applying deep learning methods also uncover hidden safety-related patterns from the physical appearance of street blocks that help address urban safety issues.

“My original intent was to improve on previous work done on crime type classification problems with spatiotemporal data such as criminal records and roadway networks,” said Fu, a Ph.D. student at the Discovery Analytics Center advised by Chang-Tien Lu. “But when we were able to access street view images from one of our sponsors, the District Department of Transportation, we saw a potential opportunity to explore crime rates and type prediction from street view images, as well.”

Next month at the 2018 ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, Fu will present a paper that he has coauthored about this research: “StreetNet: Preference Learning with Convolutional Neural Network on Urban Crime Perception.”

Fu has also worked with the District Department of Transportation (DDOT) on a research project applying social media analysis to intelligent transportation systems.

“My team and I developed a social media-based transportation status monitoring and situation summarization system for DDOT,” he said. “The proposed system monitors and retrieves transportation-related tweets and, based on the retrieved Twitter data, the system is capable of detecting traffic incidents and highlight traffic status with text summarization techniques.”

Fu has presented two coauthored papers on the research for DDOT: “Steds: Social Media Based Transportation Event Detection with Text Summarization,” at the IEEE International Conference on Intelligent Transportation Systems (ITS); and “Social media data analysis for traffic incident detection and management” at the Transportation Research Board (TRB) conference, both in 2015.

Fu holds a master’s degree in computer science from Virginia Tech and is projected to graduate with a Ph.D. in computer science in 2019.

“High influence in my area of research is mainly what attracted me to the university and the Discovery Analytics Center,” said Fu. “As such an active player in the data mining, machine learning, and urban computing research fields, DAC has provided me great opportunities for working with interdisciplinary corporations.”

DAC Student Spotlight: Xuchao Zhang

DAC Ph.D. student, Xuchao Zhang

In the era of data explosion, noise and corruption in real-world data caused by accidental outliers, transmission loss, or even adversarial data attacks is inevitable and often results in incorrect data labeling. For example, a negative review in the Internet Movie Database (IMDb) could be mislabeled as positive or an image of a panda might be mislabeled as a gibbon.

Xuchao Zhang, a Ph.D. student in computer science, is focused on solving the problem of mislabeling.

“Using scalable robust model learning, we propose distributed and online robust algorithms to handle regression and classification problems in the presence of adversarial data corruption,” said Zhang, who is advised by Chang-Tien (C.T.) Lu in the National Capital Region.

Zang said his research can be broadly applied to noisy datasets in massive real-world applications.

Zhang, who earned a bachelor’s degree at Shanghai Jiao Tong University in China, begin his Ph.D. studies in 2009.

“I chose Virginia Tech’s engineering school for its abundance of advanced research resources and outstanding faculty in the field of data mining and machine learning,” Zhang said. “I am very fortunate to work with Dr. Lu as a DAC student.”

He collaborated with Lu and other researchers from Virginia Tech and George Mason University on the study, “Online and Distributed Robust Regressions under Adversarial Data Corruption,” which he presented at the 2017 IEEE International Conference on Data Mining (ICDM) in New Orleans, LA, in November.

His research has also been presented at other conferences, including the ACM International Conference on Information and Knowledge Management (CIKM); the IEEE International Conference on Big Data, and the International Joint Conference on Artificial Intelligence (IJCAI).

Zhang serves on the program committee (research track) for the Association of Computing Machinery’s Special Interest Group on Knowledge of Discovery and Data Mining (KDD) and will be attending the 2018 conference in London.

This summer, Zhang heads to Redmond, Washington, where he has an internship at Microsoft Research AI.

DAC has strong presence at ICDM 2017

DAC Ph.D. student, Zhiqian Chen, presenting his paper at ICDM 2017.

The Discovery Analytics Center was strongly represented at the IEEE International Conference on Data Mining (ICDM) in New Orleans, Nov. 18-21, with a number of accepted research papers by DAC faculty and students and DAC faculty serving on committees and panels.

Research papers accepted for the conference include:

DAC faculty participation in the ICDM Conference included Chang-Tien Lu serving on the program committee and Naren Ramakrishnan serving as an area chair. Ramakrishnan also co-chaired a panel focusing on ethical and professional issues when dealing with social data with Tanushree (Tanu) Mitra, assistant professor of computer science, as one of the panelists. B. Aditya Prakash was invited to participate as a mentor in the ICDM Ph.D. Forum.

The ICDM has established itself as the world’s premier research conference in data mining. It provides an international forum for presentation of original research results, as well as exchange and dissemination of innovative, practical development experiences. The conference covers all aspects of data mining, including algorithms, software and systems, and applications. ICDM draws researchers and application developers from a wide range of data mining related areas such as statistics, machine learning, pattern recognition, databases and data warehousing, data visualization, knowledge-based systems, and high-performance computing. By promoting novel, high quality research findings, and innovative solutions to challenging data mining problems, the conference seeks to continuously advance the state-of-the-art in data mining. Besides the technical program, the conference features workshops, tutorials, panels.