Estimation of Photometric Redshifts. I. Machine-learning Inference for Pan-STARRS1 Galaxies Using Neural Networks

We present a new machine learning model for estimating photometric redshifts with improved accuracy galaxies in Pan-STARRS1 data release 1. Depending on the estimation range of redshifts, this based neural networks can handle difficulty inferring redshifts. Moreover, to reduce bias induced by model's ability deal difficulty, it exploits power ensemble learning. extensively examine mapping between input features and target redshift spaces which is validly applicable discover strength weaknesses trained model. Because our well calibrated, produces reliable confidence information about objects non-catastrophic estimation. While highly accurate most test examples residing space, where training samples are densely populated, its quickly diminishes sparse unobserved (i.e., unseen samples) training. report that out-of-distribution (OOD) contain both physically OOD stars quasars) observed properties not represented data. The code available at https://github.com/GooLee0123/MBRNN other uses retraining different