CS&E Announces 2023-24 Doctoral Dissertation Fellowship (DDF) Award Winners
Four Ph.D. students working with CS&E professors have been named Doctoral Dissertation Fellows for the 2023-24 school year. The Doctoral Dissertation Fellowship is a highly competitive fellowship that gives the University’s most accomplished Ph.D. candidates an opportunity to devote full-time effort to an outstanding research project by providing time to finalize and write a dissertation during the fellowship year. The award includes a stipend of $25,000, tuition for up to 14 thesis credits each semester, and subsidized health insurance through the Graduate Assistant Health Plan.
CS&E congratulates the following students on this outstanding accomplishment:
- Qiao "Georgie" Jin (Advisor: Lana Yarosh)
- Konstantinos (Costas) Mavromatis (Advisor: George Karypis)
- Dawn Michaelson (Advisors: Eric Van Wyk and Gopalan Nadathur)
- Somya Sharma (Advisors: Vipin Kumar and Snigdhansu Chatterjee)
Qiao "Georgie" Jin
Jin's thesis delved into the intersection of augmented reality (AR)/virtual reality (VR) and social computing in educational contexts. Through rigorous empirical work and the design of innovative AR/VR systems, her research provides valuable insights into the opportunities presented by these technologies and practical guidelines for developers and instructional designers to leverage them effectively. For example, in response to the AR/VR content creation barrier for instructors, her research focuses on using cost-effective materials such as 360° videos and volumetric videos to create realistic, collaborative and immersive learning environments. Specifically, her research explores the technical potential of social AR/VR as a medium to enhance education, which has been published in multiple venues and showcases her leadership in human-centered work.
Konstantinos (Costas) Mavromatis
Mavromatis’ work centers around machine learning (ML), focusing in graph machine learning (GML), graph neural networks (GNNs), and large language models (LLMs). His research is designed to address current limitations of LLMs by enhancing them with accurate domain-specific semantics provided by knowledge graphs (KG) via machine learning algorithms. Areas of impact include technological developments, such as improving the reasoning abilities of LLMs for knowledge-intensive tasks, as well as social areas, such as utilizing KGs as a tool for preventing LLMs from reproducing unjust statements. His publication record is outstanding. At the University of Minnesota, he has published five peer-reviewed conference papers, and has two papers under conference review. His peer-reviewed papers appear at the top and highly selective ML, natural language processing (NLP), and signal processing (SP) venues. In every paper, he has been the lead or equal-contributor author and representing work that he initiated and executed. Mavromatis has also published two journal papers in SP-related journals on work that he did as an undergraduate student, which is remarkable and a testament to his research drive and aspirations.
Michaelson’s thesis aims to bring together language extensibility (from Eric Van Wyk’s group), and proving languages possess mathematical properties that guarantee programs written in them are well-behaved in relevant senses (from Gopalan Nadathur’s group). This work is exciting because it raises many new theoretical challenges and has significant practical potential because both the reliability of languages and domain-specific languages are important requirements in modern day computing. Her thesis will expound ideas that will provide a strong foundation for the meta-theory of extensible programming languages and, thereby, to language development at large. Alongside this work, she has constructed a research system called Extensibella that provides a practical vindication of the ideas and supports experimentation with applications.
Climate variability has caused changes in the earth’s hydrological cycle and soil carbon cycle. Increased variability in rainfall and air temperatures is causing more drought and flood events. Similarly, changes in weather and intensive farming practices have drastically reduced soil carbon and soil water and nutrient holding capacity around the world. Due to these challenges, it is imperative to develop frameworks that can predict changes in hydrological and soil carbon cycles. This is especially useful in high-stake scenarios that can have drastic implications for human lives. This includes preventing flood or drought-related casualties in the hydrological case and crop yield failures and depletion of soil health in the soil science case.
Sharma's research focuses on improving the explainability of artificial intelligence prediction methods in the aforementioned scientific domains. As part of her dissertation, she is developing a novel physics-guided machine learning paradigm that can leverage domain knowledge in individual scientific disciplines to improve predictions and explainability of machine learning (ML) models. She is the first author of the main conference paper focusing on explainability in hydrological ML models, which was accepted to the SDM conference, a top conference in data mining. Sharma also developed a framework that can discover causal relations among soil processes under different farming and management practices. This framework is exceptional since it is a cost-effective and generalizable tool that can help farmers and agricultural businesses study the impact of their farming practices on soil nutrients and promote sustainable agricultural practices. Her work on climate change-driven crop yield failures was showcased as highlighted talks in climate change and precision agriculture workshops at top ML conferences like NeurIPS, Neural Information Processing Systems, and PAKDD, Pacific-Asia Conference on Knowledge Discovery and Data Mining.