摘要：In this talk, I present a move toward precision galaxy science by showing how uncertainties in core observables: stellar mass, star-formation histories (SFHs), halo properties, environment, and structure/morphology/kinematics, can propagate into biased inferences about galaxy evolution. For example, galaxies classified as “disks” in JWST images at z>3 may not be dynamically cold, rotationally supported systems; without robust recovery of intrinsic shape and dynamical state, such classifications can be systematically misleading. Likewise, inaccuracies in SED fitting translate directly into biased SFH and age estimates, while errors in intrinsic morphology and dynamical state can compromise studies that seek to infer feedback from observed galaxy structure and kinematics. To address these issues, I outline a Precision Galaxy System that includes generative AI to reconstruct stellar-mass distributions and (V,σ) fields from multi-band imaging at survey scale, and simulation-informed physical priors to improve SED-based inference, enabling more reliable recovery of SFHs. I then demonstrate how this framework can be applied to study galaxy assembly, star formation and quenching, and galaxy–halo connections. Finally, I discuss prospects for Euclid/LSST/CSST/SKA-era advances enabled by accurate, scalable measurements.
学术报告
From Haloes to Galaxies: Exploring Precision Galaxy Formation in the AI Era
报告题目：From Haloes to Galaxies: Exploring Precision Galaxy Formation in the AI Era
报 告 人：彭影杰 教授 北京大学
报告时间：2025-12-25 16:10:00
报告地点：天文楼212报告厅
个人简介：
Yingjie Peng is currently a Boya Distinguished Professor at KIAA, Peking University. He received his PhD from ETH Zurich in 2012, and then became a research associate at Cavendish Laboratory and the Kavli Institute for Cosmology, University of Cambridge. In 2015, he joined KIAA as a faculty member. He has been awarded the MERAC Prize from the European Astronomical Society and the National Science Fund for Distinguished Young Scholars. His research interests focus on galaxy formation and evolution, and the development of generative AI methods for astronomical inference.