Leveraging Structure to Efficiently Make Good Decisions in an Uncertain World

Опубликовано 17 августа 2016, 1:20

Making good sequential decisions under uncertainty is a core part of what it means to be intelligent. It is formally hard to make good sequential decisions when part of the environment is hidden, and most real-world problems are intractable to solve with general-purpose algorithms. Developing approaches that scale to much larger, more complex domains could have a profound impact on applications ranging from patient treatment to automated customer support to computer assisted education. In this talk I will discuss several algorithms that leverage structure common in diverse problem classes in order to efficiently make decisions in dramatically larger domains. I will present results on robot navigation over varying terrain, and I will also share results of a field study in Bangalore, India demonstrating that an adaptive tutoring software game designed for low resource schools has the potential to increase student engagement. This field trial is part of an ongoing research thrust towards improving student learning by addressing the sequential decision making challenges that arise in education.