Learning Where to Look - Reliable Certificates under Scarce Ground Truth

Explore a statistical framework for making reliable inferences when ground truth data is scarce in this academic lecture by Prof. Gautam Dasarathy from Arizona State University. Learn how to address the challenge of two-group inference under verification constraints, where revealing group membership is costly, through applications in digital biomarker validation, large language model auditing, and industrial fault detection. Discover how this approach combines active learning, sequential analysis, and classical hypothesis testing to create adaptive procedures that maintain statistical validity while improving efficiency. Examine the methodology for issuing reliable statistical certificates when only small, adaptively chosen subsets of labels are observed, and gain insights into cohort discovery techniques for uncovering subpopulations with significant effects. Understand how this framework addresses the modern challenge of making statistically sound statements about population differences when verification resources are limited across various data-driven intelligent systems.