Kai Wang, a doctoral candidate at Harvard University, will deliver a talk as part of CSRAI's Young Achievers Symposium.
“Decision-focused Learning: Integrating Optimization Problems into Training Pipeline to Resolve Social Challenges”
This talk focuses on solving social challenges formulated as optimization problems with missing parameters. For example, wildlife conservation challenges are commonly modeled as game theory problems between patrollers and poachers with unknown utility functions. Health service scheduling problems are formulated as resource allocation problems with unknown intervention effectiveness. A common way to address missing information is to learn a predictive model to predict missing parameters from domain-specific features, where actionable decisions can be obtained from solving the optimization problems with predicted parameters. However, the predictive model is trained to maximize the predictive accuracy but not the performance of the chosen decisions, leading to a mismatch between the training and evaluation objectives. Wang's research focuses on addressing the issue of mismatch objectives by expressing optimization problems, including non-convex, multi-agent, and sequential problems, as differentiable layers to integrate into the training pipeline. This novel training method leads to decision-focused learning that learns the predictive model to directly optimize the performance of the proposed decisions. Lastly, the talk concludes with experimental results in various social challenges to demonstrate the performance boost led by decision-focused learning.
About the Speaker
Kai Wang is a Ph.D. candidate studying Computer Science at Harvard University working with Professor Milind Tambe. Prior to his Ph.D., Kai graduated from National Taiwan University with a B.S. in Math and Electrical Engineering, where he won two silver medals at the International Mathematical Olympiad. Kai’s work focuses on providing actionable decisions to solve wildlife conservation and healthcare challenges. Both domains are multi-agent systems that require using machine learning to address the uncertainty involved in the system and optimization to suggest actionable solutions. Kai identifies the issue of solving machine learning and optimization problems separately, where he proposes various new techniques to integrate optimization problems into the machine learning pipeline to achieve decision-focused learning.
About the Young Achievers Symposium
The Young Achievers Symposium highlights early career researchers in diverse fields of AI for social impact. The symposium series seeks to focus on emerging research, stimulate discussions, and initiate collaborations that can advance research in artificial intelligence for societal benefit. All events in the series are free and open to the public unless otherwise noted. Penn State students, postdoctoral scholars, and faculty with an interest in socially responsible AI applications are encouraged to attend.