报告题目:Latent Tree Models and Related Fields
报告时间:2025年5月26日下午18:30
会议链接:https://meeting.tencent.com/dm/XuLPGNPOXOuE
会议 ID:593-746-752
主办单位:数学与统计学院
报告人:王晓飞
报告人简介:王晓飞,东北师范大学数学与统计学院统计系,教授,博士生导师。主要研究方向:图算法,图模型和机器学习。在国际期刊上发表多篇学术论文。主持国家自然科学基金委项目天元,青年,面上项目各一项,参与国家自然科学基金委项目多项,主持中央高校青年教师科研发展项目一项,主持吉林省优秀青年人才基金项目一项。现正与吉大一院放射科合作,研究基于国家急诊CT影像数据库的多病种精准快速联合筛查的数学方法与系统。
摘要:Latent Tree Models (LTMs) are a class of probabilistic graphical models that capture hierarchical dependencies among observed variables via unobserved (latent) variables structured in a tree topology. They serve as a powerful tool for multiscale representation, unsupervised structure discovery, and interpretable clustering. This report presents a comprehensive overview of LTMs and their related fields. We begin by revisiting the foundations of latent variable models and tree-structured probabilistic models. We then explore the learning algorithms for LTMs, including structure recovery, parameter estimation, and their theoretical guarantees.