Fall 2018 marked the first offering of SML 510 – Graduate Research Seminar. Taught by Peter Ramadge, Gordon Y.S. Wu Professor of Engineering and Professor of Electrical Engineering, the course is a semester-long series of presentations by students on their own research projects (or another topic of interest) and how machine learning fits into those efforts.
Among the nine students in the inaugural class, feedback was uniformly positive. “This was a great and successful course,” said Renzhi Jing, a fifth-year graduate student in Civil and Environmental Engineering. “I would definitely encourage my friends to enroll in the program and take this course if they are interested in stat and machine learning.”
By design, the course did not include any direct instruction in the mechanics or details of machine learning but was rather an overview of areas of research in which machine learning can play a valuable role. Each week, one student gave an hour-long presentation on his or her research. Another student introduced the speaker, and a third led a discussion after the presentation.
“Our idea was to offer graduate students a forum in which to present and discuss machine learning in the context of their own research, thus providing a wider view of the possibilities and applications of machine learning” said Prof. Ramadge.
Students in the class came from a wide range of departments, including Civil and Environmental Engineering, Computer Science, Operations Research and Financial Engineering, Mechanical and Aerospace Engineering, Physics, and Electrical Engineering.
The topics presented on were equally broad. From classic machine-learning topics like dynamic modeling and control to more speculative questions about the role of machine learning in filtering the cosmic background radiation, the presentations exposed students to a broad range of topics with machine-learning angles.
This aspect of the course was appreciated by the students. “I've learned a lot from my colleagues, not only about machine learning, but also a lot of interesting ideas in physics, neural science, finance and engineering,” said He Sun, a fifth-year student in Mechanical and Aerospace Engineering. “I cannot think of another class which can get us exposed to such wide topics.”
Other topics included modeling synthetic hurricanes, characterization of turbulent flows, short-term volatility forecasting in financial markets, predicting and downscaling agricultural water use, and neural encoding.
“It was great to meet people from across the university, and to hear about the vastly different ML problems that people are studying,” said Zoë Ashwood, a second-year graduate student in the Computer Science department.
Students also noted that the discussion and critique of their presentations were useful as they advance from graduate study toward careers. “Additionally, hearing the critique on our presentations, as well as others', was extremely useful,” said A. Stevie Bergman, a fifth-year student in Physics. “It has given me both ideas to pursue in my research, and a better sense of how machine learning is utilized across research fields.”
The small size of the class and close camaraderie of the students were also mentioned as major pluses of the course. “I feel like, for the first time since starting college, I have made attachments and created bonds and friendships with classmates and teachers,” said Average Phan, a fourth-year student in Physics. Course dinners at the middle and end of the term were especially useful in creating this environment, noted several students.
The course will be offered every semester going forward, given sufficient enrollment, and although classes always evolve, SML 510 probably won’t make any major changes any time soon, according to Prof. Ramadge.
“Of course, when we offer a new course, we make our best attempt to put together a useful and engaging class, but we don’t know until we try it whether we have succeeded,” he said. “In this case, it is gratifying that we seem to have hit our target on the first try.”