Bloomberg is excited to announce this year’s new crop of nine early-career researchers to receive the Bloomberg Data Science Ph.D. Fellowship for 2021-2022. As the program enters its fourth year, they join the three returning Fellows from 2020-2021, as the majority of Fellows from the 2018-2019 and 2019-2020 cohorts have now graduated and earned their degrees.
As part of Bloomberg’s larger efforts to engage the academic research community, the Data Science Ph.D. Fellowship awards financial aid and professional support to Fellows as they research their interests and gain real-world professional experience over the course of the academic year.
Each of these highly motivated students is engaged in research in areas relating to ongoing areas of interest at Bloomberg. In addition to providing mentorship and professional development for the next generation of data science talent, the Fellowship has helped build relationships between Bloomberg and academic researchers at a variety of institutions around the globe.
The goal of the Fellowship is to provide support and encouragement for a successful Ph.D. candidate, resulting in groundbreaking publications in both academic journals and conference proceedings, application development, open source contributions, and other forms of research dissemination. Participants were selected by a committee of Bloomberg data scientists from across the organization. They will also be supported through mentorship, career counseling, and research internships.
Let’s get to know this year’s Fellows:
University of Massachusetts Amherst
In her work in the information retrieval field, Hashemi is interested in developing machine learning models for search engines and recommender systems.
“Due to the nature of my field of research, engaging and collaborating with industry can lead to significant scientific and societal impact,” Hashemi says. “And many applications of information retrieval research exist within the Bloomberg ecosystem.”
Through her research and work at Bloomberg, Hashemi hopes to develop a framework that can lead to more accurate search and recommendations, and can also create new information access functionalities.
University of North Carolina at Chapel Hill
Having heard about the Fellowship through labmates who participated in previous years (as well as a few who have interned at Bloomberg), Zhang was eager for the opportunity to do hands-on work alongside Bloomberg’s AI research scientists to further her efforts around text summarization in the context of real-world professional challenges.
“My goal for this Fellowship is to conduct impactful research on summarization and achieve more abstractive and faithful summarization,” Zhang explains. “I would like to see my research get closer to reliable real-world applications. Summarization is an important natural language processing (NLP) task that is widely useful in many applications, for example, news summarization, meeting summarization, and document summarization. I’m interested in how to make summarization models more faithful so they can produce summaries without factual errors, hallucinations, and biases.”
Carnegie Mellon University
“My primary goal for the Fellowship is to develop effective techniques for mining and modeling dynamic networks,” Park says. “I look forward to seeing how my work can be applied to improve the way dynamic graph-structured data, such as the Bloomberg Knowledge Graph, is used in practice, while learning from that experience to further enhance the methods and identify interesting problems.”
It can be difficult to understand how real-world dynamic networks evolve over time, and to spot anomalies in them. Park’s research centers on developing tools to better approach these problems. This line of research could lead to a better understanding and use of real-world data, including temporal knowledge graphs and financial transaction networks.
School of Computing, University of Utah
With his research interpreting the meaning of text fragments and the implicit relationships between them, Gupta’s work will have a direct impact on the financial world, where semi-structured tabular data is widespread.
By working directly within Bloomberg, where large amounts of textual data from around the world are part of the day-to-day workflows, Gupta hopes to “improve tabular reasoning for semi-structured data by incorporation of on-the-fly information retrieval/extraction techniques,” and in the longer run, to look at other roadblocks in current research into understanding text in tables.
The College of Engineering and Applied Sciences at Stony Brook University/SUNY
“Studies of the financial market rely heavily on high-quality data, and Bloomberg is the information hub of the global capital markets,” says Yi, whose research centers on generating synthetic financial time series with generative adversarial networks (GANs).
Working alongside Bloomberg’s experts, Yi is excited to use his Fellowship to not only build a framework for generating financial time-series, but also to apply it to practical financial applications such as providing market forecasts and generating derivative price quotes.
USC Viterbi School of Engineering
While pre-trained language models have significantly improved over the years, Jin is hoping to take this area of data science to the next level and address longstanding problems in the field. Jin’s research focuses on allowing language models to efficiently acquire the ability to solve up-to-date tasks, such as COVID-related question answering, without performance degradation on early tasks.
“I believe news-related applications in Bloomberg could greatly inspire my research, and I am excited about the opportunity to work with highly experienced researchers,” says Jin.
While machine learning (ML) systems are currently used to improve the efficiency of reporters, data journalists, and financial advisors, Mishra argues that they often lack the domain-specific knowledge required for tackling more complex challenges.
At Bloomberg, Mishra is looking forward to taking her research on the intersection of Human-Computer Interaction and AI, and building human-centered AI tools that empower users for tasks such as information retrieval from large-scale data and human-in-the-loop AI news generation. “I cannot think of any other organization where humans are working alongside AI and having a significant impact on people’s everyday lives at such a massive scale,” says Mishra.
University of Virginia
Shi is one of a number of this year’s cohort who gravitated towards the Fellowship based on the word-of-mouth recommendation of a previous participant, Huazheng Wang (2018-2019). “His success under the support of this Fellowship motivated me to apply,” says Shi, also citing the opportunity to collaborate with top-tier research teams while exploring practical applications of his research.
Specifically, Shi’s work centers on developing efficient recommender systems, particularly within the newly proposed framework of “Federated Multi-Armed Bandits,” which incorporates features of federated learning into multi-armed bandits, a promising tool for recommender systems. Ultimately, the goal is to make this framework more suitable for real-world applications, while simultaneously improving privacy and robustness guarantees, a process Shi hopes to accelerate by seeing how data science tools are used at Bloomberg.
Carnegie Mellon University
Gupta is currently at work designing algorithms focused on distribution-free uncertainty quantification in machine learning, with two primary goals: to be theoretically verifiable and to perform in the real world.
“I am excited about getting an industry perspective on my mostly theoretical research,” says Gupta. “I am also interested in finding a principled problem to work on for my internship, where I can make practical progress, but without sacrificing statistical principles.”