Speaker
Description
I will present my current work on cosmological inference with quasars (QSOs) from the Kilo-Degree Survey (KiDS). In both the classification and the photometric redshift (photo-z) steps, I test how far the models can be extrapolated beyond the coverage of the spectroscopic training data, and how the resulting errors propagate into the cosmological constraints. I compare ML methods for uncertainty estimation in QSO photo-z, and describe the calibration of the classification and photo-z results with self-organising maps (SOMs). Finally, I present preliminary constraints from the KiDS QSO samples, obtained with tomographic binning in photo-z. Both the methods and the results carry over directly to LSST, which will deliver quasar samples that are larger and deeper, but even less well covered by spectroscopic training data.
