EECS Special Seminar: Yun-En Liu "Building behavioral experimentation engines"


Event Speaker: 

Yun-En Liu

Event Location: 

32-G449 (kiva)

Event Date/Time: 

Monday, March 9, 2015 - 4:00pm

Abstract: One of the emerging new areas of computer science is the creation of software to help people achieve their goals, such as improving health or mastering algebra. Yet identifying the right design choices is a difficult task, given that everyone has different objectives and many variables differ from person to person. It is hard to study more than a few of these variables at the same time in the lab or through multifactorial tests in online software. Nor are large observational datasets the solution, as they may not contain data about interesting or new interventions.

To tackle this problem, my work focuses on building experimentation engines attached to online software, which can automatically run experiments with the aim of improving people's lives. In this talk, I will build an overview of the components that are needed to create these engines, which require methods from many diverse fields, including HCI, machine learning, psychology, and learning sciences. I will also discuss three of my projects which contribute to this vision. First is a sampling algorithm which finds the generality of an experimental result by using tree structures that relate experimental conditions to each other, which we apply in an educational game involving number line sequences and fraction representations. Second is a sampling algorithm which finds scientifically surprising results that greatest violate our predictions, which we use to uncover surprising information about how people compute summary statistics of visual displays on Mechanical Turk. And finally, I will talk about how we can create and deploy new ways to collect data and make a difference at large scale through educational campaigns, in the form of our three Algebra Challenges and the educational game DragonBox Adaptive.




Yun-En Liu is a PhD student in Computer Science & Engineering at the University of Washington, advised by Zoran Popović (UW) and Emma Brunskill (CMU). His research revolves around the automation of experimental behavioral science, which requires creating and improving online software such as educational games, then using machine learning techniques to automatically run experiments on these systems to uncover new scientific knowledge. Yun-En received a B.A. in Computer Science from Princeton University in 2009 and an M.S. from the University of Washington in 2011. He was the recipient of an NSF Graduate Research Fellowship in 2011.