报告题目：Zero-Training Unsupervised Shallow Network for Clustering and Visualization

报 告 人：张振跃（浙江大学）

报告时间：2024年7月23日  16:00-17:00

报告地点：格物楼528

报告摘要：

Unsupervised learning plays a crucial role in various fields, providing valuable insights by reducing, grouping, and visualizing data points based on their intrinsic characteristics. This article introduces a zero-training unsuperyised shallow network (ZTSN) for high-precision clustering and visualization. We revisit graph clustering from a novel attention perspective, treating connected nodes in the adjacency graph as sharing an attention context. ZTSN employs a dynamic approach that expands the scope of attention to the entire class, eliminating the need for traditional training procedures. This zero-training methodology makes ZTSN highly versatile and applicable to a wide range of data types, including 2D-shaped data, images, texts, time series, and cutting-edge RNA-seq data. Comprehensive testing demonstrates that ZTSN provides precise clustering and highly reliable visualizations without requiring any training phase, potentially offering deep insights across various domains, including biology.

报告人简介：

张振跃，浙江大学数学学院二级教授，深圳北理莫斯科大学特聘教授。1982年2月复旦大学数学系毕业后留校任职，1989年获复旦大学博士学位，随后入职浙江大学，2013年获浙江大学心平教学杰出贡献奖。2024年入职深圳北理莫斯科大学。主要从事数值代数、科学计算、机器学习和大数据分析等研究领域的模型与算法的理论分析与计算。在国际著名学术刊物SIAM Review、SIAM J. Scientific Computing、SIAM J. Matrix Analysis and Application、SIAM J. Numerical Analysis、The Journal of American Statistical Association、IEEE Transactions on Pattern Analysis and Machine Intelligence、Journal of Machine Learning Research、Patten Recognition以及NIPS、CVPR等会议上发表高质量研究论文。张振跃教授是第一位在SIAM Review上发表研究论文的国内大陆学者。其关于非线性降维算法的工作，多年来一直列SIAM J. Scientific Computing 10年高引用率第4、5位；两个关于流形学习算法被国际机器学习中应用广泛的Scikit-Learn中收录；在相关研究中取得了许多受国际关注的基础性、系统性、理论性的研究成果。