A graduate student researching red blood cell production, another studying alternative aviation fuels, and an MBA candidate: What do they have in common? They all enrolled in 6.883/6.S083 (Modeling with Machine Learning: From Algorithms to Applications) in spring 2019. The class, offered for the first time during that term, focused on machine learning applications in engineering and the sciences, attracting students from fields ranging from biology to business to architecture.
Among them was Thalita Berpan, who was in her last term before graduating from the MIT Sloan School of Management in June. Berpan previously worked in asset management, where she observed how financial companies increasingly focus on machine learning and related technologies. “I wanted to come to business school to dive into emerging technology and get exposure to all of it,” says Berpan, who has also taken courses on blockchain and robotics. “I thought; ‘Why not take the class so I can understand the building blocks?’”
The class provided Berpan with a thorough grounding in the basics — and more. “Not only do you understand the fundamentals of machine learning, but you actually know how to use them and apply them,” she says. “It’s very satisfying to know how to build machine learning algorithms myself and know what they mean.”
Berpan plans to use what she has learned about data and algorithms to work with design engineers in her post-graduation job in project management. “What are some of the ways that engineers and data scientists can leverage a data set? For me to be able to help guide them through that process is going to be extremely useful,” she says.
Open to both undergraduate and graduate students, 6.883/6.S083 enrolled 66 students for credit in its debut semester. It was created as an experimental alternative to 6.036 (Introduction to Machine Learning), a course that professors Regina Barzilay and Tommi Jaakkola developed and initially taught, and which has become one of the most popular on campus since its introduction in 2013.
Having received feedback that 6.036 was too specialized for some non-electrical engineering and computer science (EECS) majors, Barzilay and Jaakkola designed 6.883/6.S083 to focus on different applications of machine learning. For example, Berpan, along with students from the Department of Biology and the Department of Aeronautics and Astronautics, worked on a group project that used machine learning to predict the accuracy of DNA repair in the CRISPR/Cas9 genome-editing system.
“It doesn’t necessarily mean that the class is easier. It just has a different emphasis,” says Barzilay, the Delta Electronics Professor of Electrical Engineering and Computer Science. “Our goal was to provide the students with a set of tools that would enable them to solve problems in their respective areas of specialization.”
The class includes live lectures that focus on modeling and online materials for building a shared background in machine learning methods, including tutorials for students who have less prior exposure to the subject. “We wanted to help students learn how to model and predict, and understand when they succeeded — skills that are increasingly needed across the Institute,” says Jaakkola, the Thomas Siebel Professor in EECS and the Institute for Data, Systems, and Society (IDSS).
In fact, about two-thirds of those enrolled for spring term were non-EECS majors. “We had a surprising number of people from the MIT School of Architecture and Planning. That’s very exciting,” Jaakkola says.
Ultimately, the instructors say, the new course was built to bring a variety of students together to study an exciting, fast-growing area. “They constantly hear about the wonders of AI, and this enables them to become part of it,” says Barzilay. “Obviously, it brings challenges, too, because they are now in totally new, uncharted territory. But I think they are learning a lot about their abilities to expand to new areas.”
