The University of Massachusetts Amherst
University of Massachusetts Amherst

Search Google Appliance


Travis Mandel - "Better Education Through Improved Reinforcement Learning"

DS Seminar
Monday, April 3, 2017 - 4:00pm
Computer Science Building, Room 150/151

Travis Mandel
​University of Washington

Title:  "Better Education Through Improved Reinforcement Learning"

Abstract:  Traditional approaches to education are largely based on expert intuition, as codified by textbooks and curriculum guidelines. However, this expert intuition is often limited, and furthermore, this traditional one-size-fits-all approach fails to meet the needs of a diverse population of students. With the rise of large-scale online educational platforms, we now have the potential to overcome these limitations by leveraging data collected from millions of students. Travis envisions using this data to determine the optimal content to help each student learn as quickly as possible while simultaneously remaining engaged.

This problem is a seemingly natural choice for reinforcement learning, where an AI agent learns from experience, how to make a sequence of decisions to maximize some reward signal. Although reinforcement learning has seen numerous successes in recent years, education is much more difficult in that the agent gains experience solely by interacting with actual human students. As a result, although the potential to directly impact human lives is much greater, intervening to collect new data is often expensive and potentially risky. In this talk, Travis will present methods that allow us to evaluate candidate learning approaches offline, using previously-collected data instead of actually deploying them. Further, he will present new learning algorithms which ensure that, when we do choose to deploy, the data we gather is maximally useful. Finally, Travis will explore how reinforcement learning agents should best leverage humans with some expertise (e.g. teachers) to gradually extend the capabilities of the system, a topic which lies in the exciting area of Human-in-the-Loop AI.

Throughout the presentation, Travis will discuss how he has deployed real-world experiments and has used data from thousands of kids to demonstrate the effectiveness of our techniques on educational games deployed in the wild. 

A reception will be held at 3:40 P.M. in the atrium outside the presentation room.