The Department of Statistics and Actuarial Science Spring Colloquium Series presents:

Max Sampson, Postdoctoral Research Scholar, Department of Statistics and Actuarial Science, University of Iowa

"Using Highest Conditional Density Sets to Construct Sequential Conformal Prediction Regions with Time-series Data"

Abstract: Conformal prediction is a common framework for constructing prediction sets with exchangeable data. Its use to construct prediction regions beyond exchangeability is a new and relatively underexplored area. Existing methods focus on prediction intervals with weak conditional coverage guarantees. In this talk, I will briefly review conformal prediction for exchangeable data before introducing Sequential Conformalized Density Regions (SCDR), a method for constructing prediction sets based on a conditional density estimator for time-series data. SCDR provides a doubly robust conditional coverage guarantee, which is the first of its kind to our knowledge. I will demonstrate this property via simulation before comparing SCDR with existing time-series methods on both real and simulated datasets to showcase its effectiveness. 

Meet and Greet at 3:00 pm in 241 SH. Colloquium at 3:30 pm in 61 SH.