Conference-Artile

Representative, Informative, and De-Amplifying: Requirements for Robust Bayesian Active Learning under Model Misspecification