首页

Weak gravitational lensing has served as an important probe for large-scale structure and cosmology for decades. Stage-IV surveys require both sub-percent calibration accuracy and high statistical precision for WL shear estimation, yet traditional estimators struggle with realistic galaxy complexity while machine-learning methods often introduce biases. I will introduce a physics-informed machine-learning approach that combines a fully D₄-equivariant convolutional neural network (D₄CNN) with a score-matching technique for optimal shear estimation. The D₄CNN enforces symmetry under rotations and reflections, eliminating even-order shear biases by construction, while Analytical Calibration (AnaCal) provides precise, gradient-based self-calibration.

Together with modern denoising-score-matching framework, our method achieves multiplicative biases consistent with zero at the ∼10⁻⁴ level, well within the requirement of Stage IV surveys like LSST, and reduces shape noise by ∼20% relative to the classical baseline, (equivalent to ∼40% increase in observation time), providing a principled and practical machine-learning pathway toward optimal shear estimation for Stage-IV surveys.

BIO

Shurui Lin is currently a 2nd-yr graduate student in University of Illinois, Urbana-Champaign, working with Prof.Xin Liu, after getting his bachelor’s degree of physics from University of Science and Technology of China. He now works in LSST-DESC on weak-lensing shear estimation. His research focus is applying machine learning technique for weak-lensing cosmology.

Host: Dandan Xu