CSE Lecturer Candidate Seminar #2

CSE Lecturer Candidate Seminar
Information Aggregation and Learning in Prediction Markets

Consider the following question to which a college might want an answer: Will introducing clickers in freshman science classes increase student retention? Traditionally, the answers to such questions have come from collecting data by polling the instructors and students, consulting an expert panel or surveying the college population. Prediction markets are speculative markets that are an alternative to traditional information aggregation mechanisms. They have have been shown to outperform polls and expert panels in the accuracy of their prediction. Furthermore, they respond quickly to changing data and aggregate information efficiently. However, these empirical results do not explain why these markets work as well as they do; and neither do they provide tools to inform the most effective design of such markets. One of the goals of my research has been designing prediction markets with discernible semantics of aggregation whose syntax is amenable to analysis. In my research talk, I will present a market mechanism built using a class of probability distributions called exponential families. I will explore a range of benefits arising from this construction including connections to machine learning techniques. No prior knowledge of prediction markets, machine learning or exponential family distributions will be assumed. I will also include a brief teaching demonstration on the topic of linked lists. Along the way, I will highlight my teaching philosophy and techniques that I have found to be effective modes of instruction.
Sindhu Kutty is a Visiting Assistant Professor at Swarthmore College. She obtained her Ph.D. from the University of Michigan in 2015. At Swarthmore College, she has designed and taught senior-level courses in her research area of Economics and Computation as well as Theory of Computation. At the University of Michigan, she has served as a primary instructor or Graduate Student Instructor for courses from each of the core areas of theory, AI, software, and hardware. She has received an honorable mention for the college-wide Outstanding Graduate Student Instructor Award presented by the American Society for Engineering Education. She has mentored undergraduate student research both at the University of Michigan and at Swarthmore College. Her research interests lie in the design and analysis of social computing systems, with a focus on market mechanism design and its connections to statistical machine learning.