AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-26
Complete Multi-Band Structural Characterization for 10 Million Galaxies in the Hyper Suprime-Cam Wide Survey using Deep Learning
Because galaxy structure varies with wavelength, the authors used the GAMPEN Bayesian deep-learning framework to estimate bulge-to-total light ratio, effective radius, and flux in g, r, and i bands for nearly 10 million galaxies in the HSC-SSP Wide survey. Models trained on single bands were adapted to new bands by transfer learning on about 0.5% of the data, and the posteriors were well calibrated with mean deviations under about 6%. g-band results at z ≳ 0.5 degraded significantly and were excluded from the public catalog.