题目：Some studies on robust jointly sparse soft unsupervised learning

主 讲 人：张红英 教授

时    间：2024年1月9日（周二）上午8：30

地    点：研究生自习室317

主办单位：理学院  

主讲人简介：

    张红英，西安交通大学数学与统计学院，教授，博导，主要研究领域为：粒计算，认知不确定大数据分析，统计机器学习，人工智能。已在国内外重要杂志和国际学术会议上发表学术论文50 余篇，其中， SCI 收录20余篇，ESI论文3篇，目前是中国系统工程学会模糊数学与模糊系统专委会常务委员、中国逻辑学会非经典逻辑与计算专业委员会常务委员、中国现场统计研究会统计交叉科学分会、中国人工智能学会粒计算与知识发现专委会、中国人工智能学会知识工程与分布智能专委会委员，陕西省统计学学会常务理事。多次到美国德州大学奥斯汀分校、香港中文大学、加拿大Regina 大学等国内外著名大学进行学术交流。主持国家自然科学基金面上项目多项，作为主要参与人参与科技创新2030---新一代人工智能重大项目1项。Q1杂志《Mathematics》编委。

摘要：

    As an important process of data preprocessing, unsupervised learning has received a lot of attention in the context of big data. In this talk, we'll explore three related themes. First, we will propose a kind of sparse convoluted rank PCA to handle unsupervised dimensionality reduction by considering the robustness and sparsity jointly. Second, although FCM is the most commonly used fuzzy clustering algorithm, which makes the model retain more information by extending the degree of sample belonging to the cluster to the values in the interval [0,1], it is time-consuming in processing large-scale data, and limits its application in large-scale scenarios. In addition, FCM is sensitive to noise or outliers. To solve these key problems, we integrate anchor graph and dimensionality reduction into the fuzzy clustering framework on the basis of FCM model, and effectively expand the analysis ability of fuzzy clustering algorithm in large-scale data from sample dimension and feature dimension. Furthermore, we explore the three-way space structure for clustering categorical data based on three-way concepts and we also present an initial attempt to develop sparse convoluted rank principal component analysis.