报告题目：Revisiting neural network and kernel classifiers from the perspective of Dempster-Shafer theory

从DS理论的视角重新审视神经网络及核分类器

报告人： Professor Thierry Denoeux（蒂埃里·德努教授）

报告时间：5月18日周五上午10点

报告地点：一教地下报告厅

报告摘要：

The Dempster-Shafer theory of belief function is a formal framework for modeling and reasoning with uncertainty. It is based on the representation of elementary pieces of evidence by belief functions, and on their combination by an operator called Dempster’s rule. In this talk, we show that the weighted sum and softmax operations performed in logistic regression classifiers and, for instance, in the last layer of feedforward neural networks can be interpreted in terms of evidence aggregation using Dempster's rule of combination. From that perspective, the output probabilities from such classifiers (including also support vector machines) can be seen as normalized plausibilities, for some mass functions that can be laid bare. This finding suggests that the theory of belief functions is a more general framework for classifier construction than is usually assumed, and that Dempster-Shafer theory provides a suitable framework for developing new machine learning algorithms.

DS信任函数理论是不确定性建模和推理的形式框架，其使用信任函数来表达基础的证据信息，并通过Dempster组合规则实现各条证据的组合。本场讲座将展示对于逻辑回归分类器以及前馈神经网络最终层，其中的加权求和和softmax运算可以从另一角度加以解释，也即使用Dempster组合规则进行证据的集成。此时，这类分类器（也包括支持向量机）输出的概率可以被视作归一化的似然度函数。这一发现表明，信任函数理论为构建分类器以及提出机器学习算法提供了更为通用的框架。

报告人简介： Professor Thierry Denoeux（蒂埃里·德努教授），男，法国人，1985年毕业于法国巴黎国立路桥学院，1989年获该校博士学位，1992年进入法国贡比涅技术大学任教。目前为法国贡比涅技术大学信息处理工程系教授，济南大学客座教授，国际信任函数与应用协会会长。现任International Journal of Approximate Reasoning（注）主编，IEEE Transactions on Fuzzy Systems、Applied Computational Intelligence and Soft Computing、以及Fuzzy Sets and Systems副主编，同时为多个SCI期刊的编委会成员。

研究方向包括信任函数理论、统计模式识别，不确定性建模和信息融合等。

欢迎广大师生踊跃参加！