UTOPIA - Universally Trainable Optimal Prediction Intervals Aggregation

Watch a 50-minute conference talk from the Harvard CMSA Big Data Conference where Princeton's Jianqing Fan presents a groundbreaking approach to uncertainty quantification in predictions called UTOPIA (Universally Trainable Optimal Prediction Intervals Aggregation). Learn about this novel technique that aggregates multiple prediction intervals to minimize prediction band width while maintaining coverage guarantees, applicable across biomedical science, economics, and weather forecasting. Explore how UTOPIA addresses challenges of model misspecification and sub-optimal constructions through linear and convex programming, supported by theoretical guarantees on coverage probability and optimal average length. Follow along as Fan demonstrates the method's effectiveness using synthetic data and real datasets from finance and macroeconomics, covering key concepts like conformal inference, population-level analysis, radar complexity, and VC class considerations.