In announcing the 10 newly-appointed Microsoft Ph.D. Fellows nationwide, the software company confirmed that the University of California San Diego is the only university to be awarded more than one of its Ph.D. fellowships for 2017. Each year the research division of Microsoft awards Ph.D. fellowships to a small group of particularly deserving graduate students, and the competition is intense.
Microsoft tapped two Ph.D. students from UC San Diego’s Jacobs School of Engineering for the honor. Computer Science and Engineering (CSE) graduate student Mengting Wan will use her two-year fellowship to cover all her costs for tuition, travel, and living expenses that will take Wan through to completing her Ph.D. in computer science, expected in 2019. Wan works in data mining, machine learning and computational social science, and her advisor is CSE professor Julian McAuley.
Bita Darvish Rouhani
In addition to CSE’s Wan, a second fellowship went to Bita Darvish Rouhani, a second-year Ph.D. student in the Electrical and Computer Engineering (ECE) department.
Rouhani will use her Microsoft Ph.D. Fellowship to pursue work in her chosen field of computer architecture and hardware. Her other current research interests include algorithms, machine learning and deep learning, distributed optimization, big-data analysis with low-dimensional models, reconfigurable computing and hardware/software co-design. Rouhani completed her undergraduate degree at Iran’s Sharif University of Technology in 2013, and finished an M.S. at Rice University in 2015 prior to enrolling in UC San Diego’s Jacobs School of Engineering. Since 2016, she has worked in the lab of her Ph.D. advisor, ECE Prof. Farinaz Koushanfar.
“The balance between mathematical rigor and real-world applications is the greatest strength of Mengting Wan’s research,” explained Wan’s advisor, Julian McAuley, who leads the Artificial Intelligence Group. “Her combination of strengths and interests allows her to combine models from economics, machine learning and natural-language processing, while bringing a unique perspective to explain the relationship between problems that would never have occurred to me before collaborating with her.”
Specifically, says Wan, she will use the fellowship from Microsoft to continue developing “interpretable, scalable and predictive approaches to model human opinions and behavior from structured and unstructured data.” She builds scalable, machine-learning algorithms to process massive (and heterogeneous) real-world human activity datasets, and applies it to areas including e-commerce, recommender systems (e.g., Amazon’s success in recommending future purchases based on the consumer’s previous orders on Amazon), and opinion-oriented question answering (QA) systems.
Wan models opinions and behavior in the context of two major problems. One involves “learning ‘macro-knowledge’ from ‘micro-opinions/behavior’, which aims to effectively organize and summarize both factual knowledge and subjective opinions,” she explained. “But I also want to leverage ‘macro-knowledge’ to interpret and predict ‘micro-opinions/behavior’.” To do so, Wan added, she is studying the connection between efficient recommendation engines and established behavioral theories in social sciences.
Mengting Wan is a second-year Ph.D. student in CSE, which she joined after completing her M.S. in Statistics at the University of Illinois at Urbana-Champaign in 2015. She did her undergraduate degree at Peking University, also in Statistics (receiving her B.S. in 2013).
Last summer she did an internship at the Microsoft Research facility in Redmond, WA, which may have been helpful in reinforcing her nomination for the Ph.D. fellowship.
Wan could also point to a joint paper* with her Ph.D. advisor, Julian McAuley, when she received a student travel award from the 2016 IEEE International Conference on Data Mining (ICDM) so she could present the paper in person at the ICDM meeting in Barcelona in December. She co-authored the paper on “Modeling Ambiguity, Subjectivity and Diverging Viewpoints in Opinion Question Answering Systems” with her advisor, Julian McAuley. (The paper on opinion QA systems is in contrast to the less complex task of determining the best response to a question about objective information that can be answered by constructing and exploring a factual knowledge-base). Wan and McAuley based their research on QA data from Amazon in the form of over 800,000 questions and 3,000,000 answers in eight large categories that were either “binary” or “open-ended”.
Wan is also the first author on a broad study** of “Modeling Consumer Preferences and Price Sensitivities from Large-Scale Grocery Shopping Transaction Logs.” She is headed for Perth, Australia, in April to present the paper at the 2017 International World Wide Web Conference (WWW 2017). In the paper, Wan, McAuley and their co-authors will propose a nested feature-based matrix factorization framework to model consumers’ purchase decisions from large-scale grocery shopping transaction logs.
Last year Wan received a second student travel award to present a paper at ACM Special Interest Group Conference on Knowledge Discovery and Data Mining (SIGKDD 2016) in San Francisco. It was based on her Master’s research at the University of Illinois with co-authors Xiangyu Chen, Lance Kaplan, Jiawei Han, Jing Gao and Bo Zhao – but published after Wan had left Urbana-Champaign for San Diego. The paper was titled, “From Truth Discovery to Trustworthy Opinion Discovery: An Uncertainty-Aware Quantitative Modeling Approach” (August 2016).