报告人：李钰鹏 博士

时间:  2021年12月1日（星期三）上午11:00-12:00

地点：T2-202

语言：中文

内容简介:

Malicious data manipulation reduces the effectiveness of machine learning techniques, which rely on accurate knowledge of the input data. Motivated by the real-world needs for network traffic classification, we address the problem of robust online learning in the presence of malicious data generators that attempt to gain favourable classification outcome by manipulating the data features. In this talk, we will introduce proposed online classification algorithms for the cases where the malicious generators are non-clairvoyant and clairvoyant. For each of these cases of the malicious generators, we consider both static and dynamic feedback delay over time. The proposed algorithms have theoretical performance guarantees. Our experimental results using real-world data traces demonstrate that the proposed algorithms can approach the performance of an optimal static offline classifier that is not manipulated, while outperforming the same offline classifier when tested with a mixture of normal and manipulated data. We believe our outcomes will not only inspire future research in online classification, but also have practical significance that will be conducive to entities for network operators and Internet users.

讲者简介：

Dr. Yupeng Li is an Assistant Professor at Hong Kong Baptist University. He was a post-doctoral researcher at University of Toronto. His research interests are in general areas of network science and, in particular, algorithmic decision making and machine learning problems, which arise in networked systems such as information networks, social networks, the edge-clouds, and transportation networks. He is also excited about interdisciplinary research that applies algorithmic techniques to edging problems. He has served as TPC member and reviewer in some top-level international conferences and journals, and he has published papers in prestigious venues such as ACM MobiHoc, IEEE INFOCOM, IEEE Journal on Selected Areas in Communications, and IEEE/ACM Transactions on Networking.