报告人简介：

专长和学术成就：算法设计，计算几何，组合优化，及他们在医学图像，治疗规划，及诊断 ，生物 ， 网络与移动计算，大规模集成电路设计 等方面的应用。在这些方向的国际期刊和会议上发表了140余篇论文，大部分出现在国际顶级期刊和会议中。解决了几个长期公开的问题和猜想。推广了一类最基本的几何结构和经典问题。

工作单位及主要简历：于1992年和1995获得中国科技大学，计算机科学本科和硕士学位。于2000年获得美国圣母大学（University of Notre Dame）计算机科学与工程博士学位。同年获美国纽约州立大学（布法罗）计算机科学与工程系助理教授职位，现为该系终身正教授。

1. 报告题目: Finding Global Optimum for Truth Discovery: Entropy Based Geometric Variance

  报告时间：2015年6月30日下午14:00

  报告地点：理工楼555

Abstract：Truth Discovery is an important problem arising in data analytics related fields such as data mining, database, and big data. It concerns about finding the most trustworthy information from a dataset acquired from a number of unreliable sources. Due to its importance, the problem has been extensively studied in recent years and a number techniques have already been proposed. However, all of them are of heuristic nature and do not have any quality guarantee. In this paper, we formulate the problem as a high dimensional geometric optimization problem, called Entropy based Geometric Variance. Relying on a number of novel geometric techniques (such as Log-Partition and Modified Simplex Lemma), we further discover new insights to this problem. We show, for the first time, that the truth discovery problem can be solved with guaranteed quality of solution. Particularly, we show that it is possible to achieve a (1+eps)-approximation within nearly linear time under some reasonable assumptions. We expect that our algorithm will be useful for other data related applications.

2.报告题目: Clustering-Based Collaborative Filtering for Link Prediction

  报告时间：2015年7月3日下午14:00

  报告地点：理工楼555

Abstract：In this paper, we propose a novel collaborative ?ltering approach for predicting the unobserved links in a network (or graph) with both topological and node features. Our approach improves the well-known compressed sensing based matrix completion method by introducing a new multiple independent-Bernoulli-distribution model as the data sampling mask. It makes better link predictions since the model is more general and better matches the data distributions in many real-world networks, such as social networks like Facebook. As a result, a satisfying stability of the prediction can be guaranteed. To obtain an accurate multiple-independent Bernoulli-distribution model of the topological feature space, our approach adjusts the sampling of the adjacency matrix of the network (or graph) using the clustering information in the node feature space. This yields a better performance than those methods which simply combine the two types of features. Experimental results on several benchmark datasets suggest that our approach outperforms the best existing link prediction methods.

3.报告题目: Random Gradient Descent Tree: A Combinatorial Approach for SVM with Outliers

  报告时间：2015年7月5日下午14:00

  报告地点：理工楼555

Abstract：Support Vector Machine (SVM) is a fundamental technique in machine learning. A long time challenge facing SVM is how to deal with outliers (caused by mislabeling), as they could make the classes in SVM nonseparable. Existing techniques, such as soft margin SVM, ν-SVM, and Core-SVM, can alleviate the problem to certain extent, but cannot completely resolve the issue. Recently, there are also techniques available for explicit outlier removal. But they suffer from high time complexity and cannot guarantee quality of solution. In this paper, we present a new combinatorial approach, called Random Gradient Descent Tree (or RGD-tree), to explicitly deal with outliers; this results in a new algorithm called RGD-SVM. Our technique yields provably good solution and can be ef?ciently implemented for practical purpose. The time and space complexities of our approach only linearly depend on the input size and the dimensionality of the space, which are signi?cantly better than existing ones. Experiments on benchmark datasets suggest that our technique considerably outperforms several popular techniques in most of the cases.