报告人：杨圣祥教授
工作单位：德蒙福特大学（英国）
报告题目：标签稀缺下的动态数据流挖掘
报告时间：2022年10月4日（周二）15:00-16:00
报告链接：https://meeting.tencent.com/dm/RaTpV6dTF4O1（腾讯会议）
（会议号：218-552-049）
内容摘要：
数据流挖掘是传统数据挖掘的一个自然且必要的发展历程。然而，这对批量分析增加了新的挑战：除了严格的时间和内存限制外，变化是一个主要因素。在一个动态的数据流中，底层概念可能会随着时间的推移发生漂移或变化，识别该变化并作出反应在标签稀缺的情况下变得更加复杂。本讲座将介绍我们最近的工作，评估无监督学习作为标签稀缺情况下动态数据流中在线分类的基础。并介绍一种基于蚁群行为的新型流聚类算法，称为蚁群流聚类（ACSC）。此外，还将介绍一个新的框架，即聚类和分类的集成学习（COCEL），用于在标签稀缺的动态流中进行分类。所提出的框架能够识别和应对数据流中的变化，并能极大地减少所需标签的数量（通常低于整个数据流的0.05%）。最后，将作出一些结论。
个人简介：
杨圣祥教授现任英国德蒙福特大学（De Montfort University）计算机科学与信息学院教授和人工智能研究院副主任。杨圣祥教授长期从事计算智能理论、方法及应用研究，在计算智能方法、进化计算求解动态优化问题、智能网络优化等方面的研究做出了突出贡献，其研究工作得到英国工程和物理科学基金会、英国皇家工程学会、英国皇家学会、欧盟以及工业界的大力资助，先后承担了30余项科研基金项目，出版英文编著2部，编辑国际会议论文集8部, 发表论文370多篇，其中SCI期刊论文160余篇，其Google Scholar 引用15500余次，H-index为64。杨圣祥教授入选美国斯坦福大学发布的全球前 2% 顶尖科学家榜单 (World's Top 2% Scientists 2021)，并应邀担任10余种国际知名期刊（包括《IEEE Transactions on Evolutionary Computation》和《IEEE Transactions on Cybernetics》）的副主编或编委，担任国际大会程序委员会主席和分会主席50余次，应邀做国际会议大会报告或专题报告30余次。
【编辑：王健】
英文版：
Speaker: Professor Shengxiang Yang
Title: Dynamic Data Stream Mining with the Scarcity of Labels
Time: 15:00-16:00, Octoober 4, 2022 (Tuesday)
Website: https://meeting.tencent.com/dm/RaTpV6dTF4O1 (Tencent Meeting)
(Meeting number:218-552-049)
Abstract:
As pointed out in some classical papers on control, the extension from one-degree-of-freedom (1-DOF) controllers to two-degree-of-freedom (2-DOF) controllers enables the separate design with respect to the reference input (or the set-point) and the disturbance input (usually of load- type) in terms of a feedforward connection and transfer element inserted in the controller (and thus control system) structure. This allows the design of control systems with good dynamics performance with respect to both the reference input and the disturbance input, namely good set-point tracking and good disturbance rejection. The transition of the results specific to linear controllers to the fuzzy ones was suggested by Precup and Preitl in 1999 and 2003 leading to 2-DOF fuzzy controllers, which were first called fuzzy controlllers with non- homogenous dynamics with respect to the input channels, and further developed in 2009 and 2012 and applied to servo systems and electrical drives. This transition gives the opportunity to improve the control system performance especially when dealing with nonlinear processes.This lecture presents several issues concerning the design, tuning and implementation of 2-DOF fuzzy controllers focusing on 2-DOF PI-fuzzy controllers and 2-DOF PID-fuzzy controllers in their Mamdani and Takagi-Sugeno-Kang forms. The tuning is based on mapping the parameters of the linear PI and PID controllers to the parameters of the fuzzy controllers in terms of the modal equivalence principle. The linear controllers are tuned by Preitl’s and Precup’s Extended Symmetrical Optimum method (1999). The classical algebraic approach based on Diophantine equations and mapping the parameters of the 2-DOF linear controllers to the parameters of the 2-DOF fuzzy ones will be treated as well. This lecture highlights a part of the results obtained by the Process Control group in applications of 2-DOF fuzzy controllers. The results outlined in this lecture are related to processes in representative lab equipment in Process Control group’s labs and control systems in past and ongoing research contracts. Digital simulation results and experimental results are included.
[Editor: Jian Wang]