The Prediction, Inference, and Scale in Data Science Models (PRISM) Lab explores the theoretical foundations of prediction and inference in data science, with a focus on how model scale—large or small—shapes statistical properties and decision-making. We study why models optimized for prediction often diverge from those suited for inference, and examine this behavior in both overparameterized and underparameterized settings. We bridge theory and practice by investigating when high-variance prediction machines outperform more stable but less accurate alternatives, and how information is leveraged when overparameterized models rely on (implicit) regularization.
Research Tags: Prediction; Inference; Large-scale models; Biomedical science; Public Health Science; Diagnostic Imaging; Diagnostic Trials; Statistical Models; Biostatistics
