A new formulation for likelihood-informed Bayesian inference is proposed in this work based on probability models introduced for the features between the measurements and model predictions. The formulation applies to both ...