Kyu Chang’s thesis, aptly titled “Explaining Active Learning Queries,” explored ways to explain algorithm-produced data for chemists as a part of the Dark Reactions Project.
The Dark Reactions Project (DRP) is a collaboration between professors in Haverford’s chemistry and computer science departments that are using unpublished “dark” (or unsuccessful) reaction to create a machine-learning algorithm that is able to predict chemical reaction successes or failures with greater accuracy than human intuition.
In some cases, the DRP algorithm predictions for recommended reactions are hard to understand and the algorithm lacks a method for explaining its predictions. The computer science major and statistics minor from Seoul, South Korea hoped to incorporate an explanatory procedure into the algorithm for chemists to better understand its predictions.
“The biggest takeaway from the project is the research experience,” Chang says, “Research is very focused on one subject and is open-ended. The process of approaching an unsolved problem was an invaluable experience for me.”
In late July, Chang will be headed to the Googleplex in Mountain View, Calif. where he will begin working as a software engineer.
What are the implications for your thesis research?
The Dark Reactions Project aims to help chemists collect better data in a more efficient way by utilizing machine-learning algorithms. In the Dark Reactions Project we have been using a technique called “active learning” to recommend reactions to chemists. One problem the project encountered was a lack of explanation for the recommended reactions. My thesis was about providing an easy-to-interpret explanation for data instance chosen by an active learning algorithm. The explanations are provided to a chemist, who is able to interpret the explanations that are helpful in identifying chemical space that is poorly understood by the model. The technique I developed will be improved upon and studied further by other Haverford students and my advisor, Sorelle Friedler.
Is there anything else I should know about you, your work, or your time at Haverford?
When I was a freshman I thought I would major in economics. After my first year at Haverford, I served in the Korean military for two years. After the service, I returned to Haverford as a sophomore, and it was then when I took an introductory computer science course with Professor David Wonnacott. I really enjoyed the course and taking it led me to major in computer science. I definitely benefitted from Haverford’s academic flexibility, and I’m not sure if I could make such change (from economics to computer science) easily if I didn’t attend a liberal arts college.
Photo courtesy Argonne National Laboratory.
“What They Learned” is a blog series exploring the thesis work of recent graduates.