时间：2017年5月22日（星期一）14:00 - 16:00
周冰博士，现任教于美国萨姆休斯顿州立大学计算机系，于2012年在加拿大里贾那大学获得计算机博士学位。研究方向包括粗糙集，粒计算，数据挖掘，机器学习，信息安全等。 发表多篇SCI/EI检索论文，有近600次的论文引用率；获得国际粗糙集会议最佳论文奖2次，4篇论文被评为Web of Science前1%的高引用率文章。担任多个杂志的审稿人，多个国际会议程序委员会委员，组织过多个Workshop并担任主席。
The talk has three parts:
1. Two Bayesian Approaches to Rough Sets
Bayesian inference and probabilistic rough sets provide two methods for data analysis. Both of them use probabilities to express uncertainties and knowledge in data and to make inference about data. Many proposals have been made to combine Bayesian inference and rough sets. A uniﬁed framework is presented that enables us a) to review and classify Bayesian approaches to rough sets, b) to give proper perspectives of existing studies, and c) to examine basic ingredients and fundamental issues of Bayesian approaches to rough sets. By reviewing existing studies, we identify two classes of Bayesian approaches to probabilistic rough sets and three fundamental issues. One class is interpreted as Bayesian classiﬁcation rough sets, which is built from decision-theoretic rough set models proposed by Yao, Wong and Lingras. The other class is interpreted as Bayesian conﬁrmation rough sets, which is built from parameterized rough set models proposed by Greco, Matarazzo and S lowin´ski. Although the two classes share many similarities in terms of making use of Bayes’ theorem and a pair of thresholds to produce three regions, their semantic interpretations and, hence, intended applications are dirough set models give a practical technique for estimating probability based on Bayes’ theorem and inference. Finally, a theory of three-way decisions oﬀers a tool for building ternary classiﬁers.
2. Decision-Level Sensor-Fusion based on DTRS
A decision-level sensor fusion based on decision-theoretic rough set (DTRS) model is proposed. Sensor fusion is the process of combining sensor readings from disparate resources such that the resulting information is more accurate and complete. Decision-level sensor fusion combines the detection results instead of raw data of diﬀerent sensors, and it is most suitable when we have diﬀerent types of sensors. Rough set theory oﬀers a three-way decision approach to combine sensor results into three regions and reasoning under uncertain circumstances. Based on DTRS, we build a cost-sensitive sensor fusion model. A loss function is interpreted as the costs of making diﬀerent classiﬁcation decisions, the computation of required thresholds to deﬁne the three regions is based on the loss functions. Finally, an illustrative example demonstrates the framework’s eﬀectiveness and validity.
3. Utilizing DTRS for Imbalanced Text Classiﬁcation
Imbalanced data classiﬁcation is one of the challenging problems in data mining and machine learning research. The traditional classiﬁcation algorithms are often biased towards the majority class when learning from imbalanced data. Much work have been proposed to address this problem, including data re-sampling, algorithm modiﬁcation, and cost-sensitive learning. However, most of them focus on one of these techniques. We propose to utilize both algorithm modiﬁcation and cost-sensitive learning based on decision-theoretic rough set (DTRS) model. In particular, we use naive Bayes classiﬁer as the base classiﬁer and modify it for imbalanced learning. For cost-sensitive learning, we adopt the systematic method from DTRS to derive required thresholds that have the minimum decision cost. Our experimental results on three well-known text classiﬁcation databases show that uniﬁed DTRS provides similar performance on balanced class distribution, outperforms naive Bayes classiﬁer on imbalanced datasets, and is competitive with other imbalanced learning classiﬁer.