Kai Wang, Harvard University

Abstract:

Decision making under uncertainty is a fundamental challenge in the area of artificial intelligence for social impact. My research focuses on combining machine learning to tackle uncertainty and decision-making processes to suggest actionable solutions. First, I study integrating decision-making processes as a differentiable layer in the machine learning pipeline to achieve end-to-end learning. I identify the scalability challenge and generalize the existing integration to a broader range of planning processes involved in social challenges, including non-convex, sequential, and multi-agent planning problems accompanied by new learning methods to reduce the computation cost. Second, I incorporate decision-making processes to guide online learning decisions. My work shows that domain knowledge in planning processes provides additional information to shrink uncertainty and improve exploration performance. Lastly, I partner with NGOs in maternal and child health and pollution control to tackle uncertainty, suggest actionable solutions, and convert my work to social impact.