Title: CoLFI: A Framework for Likelihood-free Inference with Neural Density Estimators

Speaker: 王国建

Institute: 南非夸祖鲁-纳塔尔大学

Host: 顾为民

Time: 16:30-17:30, Thursday, March, 23

Location: Physics Building 552

Abstract:

In this talk, I will present how to constrain cosmological parameters with Neural Density Estimators (NDEs). Specifically, I will show three NDEs that we presented: the artificial neural network (ANN), the mixture density network (MDN), and the mixture neural network (MNN). I will show the principle of estimating parameters using these three NDEs. Then I will show a flexible code called CoLFI that we developed to constrain parameters using ANN, MDN, and MNN, which is suitable for any parameter estimation of complicated models in a wide range of scientific fields. CoLFI can obtain a high-fidelity posterior distribution using O(10^2) simulation samples, which makes parameter estimation faster, especially for complex and resource-consuming cosmological models. In addition, I will briefly introduce the 21 cm related research I participated in, and look forward to the application prospects of CoLFI and machine learning methods in 21cm cosmology.

Bio：

王国建，2015年于云南大学物理系获学士学位，2020年于北京师范大学天文系获博士学位，同年在北京师范大学天文系担任科研助理，2021年至今在南非夸祖鲁-纳塔尔大学从事博士后研究。他从事观测宇宙学研究，目前主要侧重于探索机器学习在宇宙学中的潜在应用，21 cm 相关的数据处理，以及探索人工智能技术用于宇宙学和天体物理学的发现。已在 Nature communications，美国天体物理学杂志(ApJ)，美国天体物理学杂志增刊(ApJS), 英国皇家天文学会月刊(MNRAS)发表SCI论文13篇，h-index 为 10。他目前是美国-南非HERA望远镜的成员。他参与国家自然科学基金天文联合基金重点支持项目1项。目前是ApJ和Research in Astronomy and Astrophysics的审稿人。