CosmologyAnnotated2023
Zephyr : Stitching Heterogeneous Training Data with Normalizing Flow for Photometric Redshift Inference
Z. Sun, J. Speagle, S. Huang, Y.-S. Ting, & Z. Cai
Neural Information Processing Systems (NeurIPS) Workshop 2023
Show overview ▸Hide overview ▾
Zephyr uses normalizing flows to combine heterogeneous reference-redshift samples, propagate photometric uncertainty and expose each training component’s contribution. On the reported test set it lowers scatter and outlier rates relative to two baselines; reference-data limitations and approximate uncertainty treatment remain relevant.
Open PDFLoading page…

Annotify adds explanatory annotations and fades the original page background. Source paper: CC BY 4.0.
Open annotated PDF