首页

基于图神经网络的统一局部离群点异常检测方法

发布日期:2022-12-19      新闻来源:       责任编辑:沙艳      地点:腾讯会议(ID: 678-713-690)                日期:2022.12.21

学术报告-Unifying Local Outlier Detection Methods via Graph Neural Networks

报告题目:Unifying Local Outlier Detection Methods via Graph Neural Networks

报告专家:See-Kiong Ng(黄思强)新加坡国立大学计算学院教授

报告时间:20221221 上午10:00

报告地点:腾讯会议(ID: 678-713-690)

主办单位:必赢626net入口 必赢626net入口

欢迎广大师生参加!

Abstract:

As data grows in abundance and importance for decision-making in many industries, anomaly detection methods that are fast, accurate and automated are increasingly vital. Many well-established anomaly detection methods use the distance of a sample to those in its local neighbourhood: so-called `local outlier methods', such as LOF and DBSCAN. They are popular for their simple principles and strong performance on unstructured, feature-based data that is commonplace in many practical applications. However, they cannot learn to adapt for a particular set of data due to their lack of trainable parameters. In this work, we begin by unifying local outlier methods by showing that they are particular cases of the more general message passing framework used in graph neural networks. This allows us to introduce learnability into local outlier methods, in the form of a neural network, for greater flexibility and expressivity: specifically, we propose LUNAR, a novel, graph neural network-based anomaly detection method. LUNAR learns to use information from the nearest neighbours of each node in a trainable way to find anomalies. We show that our method performs significantly better than existing local outlier methods, as well as state-of-the-art deep baselines. We also show that the performance of our method is much more robust to different settings of the local neighbourhood size.


Speaker bio:

 See-Kiong Ng (Ph.D. Carnegie Mellon University) is a Professor of Practice at the Department of Computer Science of the School of Computing at National University of Singapore (NUS), and the Deputy Director of the university’s Institute of Data Science. See-Kiong obtained his Bachelor (1989), Masters (1993), and Ph.D. (1998) degrees in Computer Science from Carnegie Mellon University and a Masters (1991) degree in Artificial Intelligence from University of Pennsylvania. Prior to joining NUS in 2016, See-Kiong was a Programme Director of the Urban Systems Initiative by the Science and Engineering Research Council of the Agency of Science, Technology and Research (A*STAR), and the founding head and principle scientist of its Data Mining Department. Currently, See-Kiong’s mission at NUS is to leverage his data science and AI expertise for transdisciplinary and translational research into important real-life problems and education of the next generation of data scientists. From using the computation of data to better understand the biology of the human body, See-Kiong is using machine learning and artificial intelligence to understand the “biology” of complex human cities and societies and creating real-world impact with the science of data.


讲题:基于图神经网络的统一局部离群点异常检测方法

讲座大纲

随着数据的日益丰富和对许多行业决策的重要性,快速、准确和自动化的异常检测方法越来越重要。许多成熟的异常检测方法使用样本与其局部邻域内样本的距离,即所谓的"局部离群点方法",如LOFDBSCAN。它们以其简单的原理和对许多实际应用中常见的非结构化、基于特征的数据的强大性能而备受欢迎。然而,由于缺乏可训练的参数,它们无法学习适应特定的数据集。在这项工作中,我们首先统一了局部离群点方法,表明它们是更一般的图神经网络消息传递框架的特例。这使得我们可以将可学习性以神经网络的形式引入到局部离群点方法中,以获得更大的灵活性和表达能力。具体来说,我们提出了一种新的基于图神经网络的异常检测方法LUNARLUNAR学习以可训练的方式使用每个节点的最近邻居的信息来发现异常。实验表明LUNAR的性能显著优于现有的局部离群点方法,以及最先进的深度学习方法。此外,LUNAR的性能对局部邻域大小的不同设置更加稳健。

主讲人简介:

黄思强博士为新加坡国立大学计算学院计算机科学系教授,新加坡国立大学数据科学研究所副所长。1986年,他获得新加坡国家电脑局海外奖学金赴美国学习,先后获得卡内基梅隆大学的计算机科学学士(1989年)、硕士(1993年)和博士(1998年)学位,以及宾夕法尼亚大学的人工智能硕士学位(1991年)。黄博士曾担任新加坡科技研究局(A*STAR)组织的城市体系计划主任,并担任了新科研信息通信研究所数据分析部门的创始负责人和首席科学家。目前,黄博士于国大新设立的数据科学研究所,着力为新加坡培养新一代杰出的数据科学家,并与多个行业与公共机构进行成功的研究合作,以数据科学开发实用的人工智能技术。

作为一名经验丰富的数据科学家,从使用数据挖掘和机器学习来揭示人体的生物学(生物信息学),到使用大数据和人工智能来理解复杂的人类城市(智能城市),黄博士以数据的科学与应用证明了数据中的巨大价值。