Title: Cosmological Forensics: Reading the Geometry of Parameter Inference
Speaker: Nikolina Sarcevic (Duke)
Abstract:
What features of a cosmological model determine the shape of its posterior, and how does that shape change when the data depart from our expectations? I will use derivative-based approximations to explore the local geometry of parameter inference, from familiar Fisher forecasts to higher-order descriptions of non-Gaussian structure. I will also discuss computational tools for constructing these approximations and emerging applications that use posterior geometry to diagnose the behavior of an analysis.
