"Again! – But Faster, Better, and with More Physics: ML-Accelerated Inference of Galaxy Properties in Deep and Wide Surveys of the Universe"

The inference of physical properties of galaxies at cosmological distance requires modeling a wide range of physics, including stellar evolution and atmospheres, dust attenuation and re-emission, nebular physics, AGN emission, and more. Bayesian inference is typically used to map the inevitable degeneracies, but current codes are computationally expensive — a significant challenge given that current and near-future surveys will yield spectra for millions of galaxies and imaging for billions. New approaches introduced in this talk include neural net emulators of key physics (photoionization modeling, stellar spectra), efficient gradient-enhanced GPU-accelerated high-dimensional sampling, and rapid simulation-based inference. These methods enable speed-ups of 100x to 1,000,000x with different trade-offs in flexibility and accuracy. New science directions include modeling entire galaxy populations simultaneously and extremely high-dimensional spatially resolved modeling of individual systems. The talk will be recorded and uploaded to the IAU Commission J1 YouTube Channel.