Post Selection

Post-selection inference (PoSI) addresses the challenge of drawing valid statistical conclusions after a model or subset of data has been selected based on the same data used for inference. Current research focuses on developing computationally efficient algorithms, such as those leveraging parametric distributions or variational methods, to construct valid confidence intervals and p-values in various settings, including online prediction and constrained sampling. These advancements improve the reliability of inferences made after model selection, impacting fields like machine learning where model selection is ubiquitous, and ensuring more robust conclusions in scientific studies.