报告题目：C3NN: Fast Extraction of N-Point Information with Physics-Informed Neural Networks and its applications to cosmological large-scale structure 

报告人：Zhengyangguang Gong (龚郑阳光) 博士（University of Arizona）

报告摘要：Modern cosmological research on large-scale structure has seen a rapid rise in the use of machine learning (ML) methods. Among these,Convolutional Neural Networks (CNNs) have gained particular attention for their remarkable performance in image classification,cosmological parameter inference,and related applications. However, many CNN-based models are often criticized as“black boxes”，as their learned summary statistics lack interpretability. Our approach, the Cosmological Correlator Convolutional Neural Network (C3NN), addresses this challenge by combining the efficiency of ML-based information extraction with a transparent interpretation of the resulting summary statistics. By imposing specific constraints on a single-layer CNN architecture, we analytically demonstrate that its summary statistic corresponds to an integral over a real-space n-point correlation function of a given order. This framework effectively "opens the black box"，enabling us to quantitatively rank the contributions of interpretable convolutional outputs to classification tasks and to study the information content relevant for cosmological parameter inference—particularly for n-point functions that are otherwise computationally prohibitive to evaluate in modern galaxy and weak lensing surveys. 

报告时间：2025年10月28日（周二）10:00

报告地点：仙林园区3-402室

欢迎大家参加！