|Time||to 01:00 pm Add to Calendar 2023-03-31 12:00:00 2023-03-31 13:00:00 HPC hybrid Seminar Friday March 31st on Trustworthy AI 3505 N. Charles Street & Zoom Population Research Institute America/New_York public|
|Location||3505 N. Charles Street & Zoom|
|Presenter(s)||Dr. Anqi Liu, Assistant Professor of Computer Science|
"A Conservative Extrapolation Approach to Trustworthy AI"
The unprecedented prediction accuracy of modern machine learning beckons for application in a wide range of real-world situations, including healthcare, education, and hiring. A key challenge is the difficulty to collect data from diverse enough populations. It causes a well-known problem called distribution shift, which means the test cases are not well-represented by the training data and usually leads to inequivalent performance in different subgroups. Overconfident errors in the under-represented group brings harm to social trust in AI-based services. In these cases, we must go beyond the conventional learning paradigm of maximizing average prediction accuracy with generalization guarantees that rely on strong distributional relationships between training and test examples. In this talk, I will describe a distributionally robust learning framework that offers rigorous guarantees under data distribution shift. This framework yields appropriately conservative extrapolations and can be used for producing more equitable prediction results among subgroups. I will also introduce a survey of other real-world applications that would benefit from this framework for future work.