12:00–1:00 pm Zoom
Josh Speagle (University of Toronto) "Cosmological Cartography with Photometric Redshifts"
The next generation of large-area surveys such as LSST/Rubin and Euclid aim to constrain cosmological parameters through weak lensing with photometric datasets comprising >1B objects. Deriving accurate, reliable, and robust photometric distances (redshifts; "photo-z's") and uncertainties to these objects remains one of the dominant systematics in current analyses (e.g., DES, HSC, KiDS). I will first talk about efforts to develop a simple, interpretable, and astronomy-oriented machine learning framework for trying to estimate photo-z's drawing on a combination of supervised and unsupervised methods. I will then describe how we are able to use this approach to both flag and remove outliers, as well as preview more recent work trying to characterize outlier populations "on the fly". I will then discuss how leveraging this framework allowed for novel systematics tests for weak lensing analyses. I will close by highlighting opportunities to exploit hierarchical inference to better propagate photo-z and other uncertainties into downstream cosmological analyses.