报告题目：New Models for Data Analysis Based on Nonlinear Integrals

报告人：王震源 教授

报告时间：2014年10月19日（周日）下午3点

报告地点：信息馆401学术报告厅

报告人简介：王震源1962年毕业于复旦大学数学系，1991年获美国纽约州立大学(Binghamton)博士学位。

从1962年开始，王震源任教于河北大学。1979至1981年，以访问学者身份在法国巴黎第六大学概率计算实验室和人工智能实验室从事非可加测度和非线性积分的研究。回河北大学后，历任副教授（1983-1986）、教授（1986-2000）、数学系系主任（1985-1990）。自1989年起，先后在美国宾厄姆顿大学(SUNY) 系统科学系、新墨西哥州立大学数学系、得克萨斯大学 (El Paso) 数学系、以及香港中文大学计算机科学和工程学系分别任客座教授/研究员。王震源自2001年起任教于美国内布拉斯加大学(Omaha)，现为该校数学系终身教授。

王震源曾获河北省科技进步一等奖(1985)、国家科委和劳动人事部颁发的国家级具有突出贡献的中青年科技专家称号(1986)、ISI (美国科学信息研究院, SCI发布者)的经典引文奖(2000)、美国内布拉斯加大学杰出研究和创造性工作奖(2007)等奖励和荣誉称号。他已发表科学论文一百五十余篇，并出版三部专著：Fuzzy Measure Theory（Plenum，1992）、Generalized Measure Theory （Springer, 2008）、Nonlinear Integrals and Their Applications in Data Mining （World Scientific，2010）。他是Fuzzy Sets and Systems等四个国际杂志的编委或副主编。王震源曾任第七、八、九届全国政协委员。

报告简介：In information fusion, regarding the set of considered predictive attributes (in classification, called feature attributes) in a data base as the universal set, nonadditive set functions defined on its power set can effectively describe the interaction among the contribution rates from various predictive attributes towards the fusing target, which can be regarded as a specified objective attribute. Such type of interaction is totally different from the traditional statistical correlationship. Relevantly, the classical linear aggregation tool, weighted sum, which can be expressed as a linear integral defined on the universal set, should be generalized to be some nonlinear integral. The Choquet integral, the upper integral, and the lower integral are the common types of nonlinear integrals. Data mining is just an inverse problem of information fusion. Using nonlinear integrals, some classical models in data mining, such as the multiregression and the classification, can be generalized as well. Once the necessary data set is available, the values of unknown parameters in these nonlinear models can be optimally determined through some soft computing techniques, including genetic algorism and pseudo gradient search, approximately. Since the above mentioned interaction can be elaborately captured, the introduced new nonlinear models are significant and powerful in practice. They may be widely applied in bioinformatics, medical statistics, economics, forecast, decision making et al. In face of various challenges from big data, these nonlinear models may have relevant generalizations, adjustments, improvements, and deformations.

欢迎广大师生参加！



信息科学与工程学院

2014年10月15日