Description
Detecting Modeling Bias at the field level: Applications to HSC Y3
Simulation-based inference provides a powerful framework for extracting rich information from nonlinear scales in current and upcoming cosmological surveys, and ensuring its robustness requires stringent validation of forward models. In this talk, I frame forward model validation as an out-of-distribution (OoD) detection problem, where the field-level probability density serves as a diagnostic tool—analogous to a chi-squared test but applied at the field level. Using weak lensing maps, I demonstrate that the field-level likelihood density effectively identifies systematic modeling errors, such as baryonic feedback, and significantly outperforms summary statistics like the scattering transform (ST) or convolutional neural network (CNN)-learned statistics. Finally, I apply this framework to the HSC Y3 data to assess forward model validity and enhance simulation-based inference.
