20210317-06 总第231期 深度学习中的拓扑美学：图神经网络

报告嘉宾：黄增峰（复旦大学）

报告题目：Embedding Methods From Graph Partition Algorithms and Their Connection to GNNs

报告嘉宾：王啸（北京邮电大学）

报告题目：Dive into the Message Passing Mechanism of Graph Neural Networks

Panel嘉宾：

黄增峰 (复旦大学)、王啸 (北京邮电大学)、何向南 (中国科学技术大学)、崔鹏 (清华大学)、Jian Tang (Mila (Quebec AI institute) and HEC Montreal)

Panel议题：

1. 如何看待当前图上的自监督学习进展？未来前景如何？

2. 图神经网络当前的落地场景？图神经网络在实际落地中还需要考虑哪些因素？

3. 图神经网络可能应用于哪些交叉学科（如生物医学工程、材料基因组等）？

4. 当下，图神经网络有哪些问题亟待解决，哪些研究方向值得关注？

5. 对想研究图神经网络的新入门同学有何建议？

6. 能否展望一下图神经网络在未来几年的发展趋势？

*欢迎大家在下方留言提出主题相关问题，主持人和panel嘉宾会从中选择若干热度高的问题加入panel议题！

报告嘉宾：黄增峰 (复旦大学)

报告时间：2021年3月17日 (星期三)晚上20:00 (北京时间)

报告题目：Embedding Methods From Graph Partition Algorithms and Their Connection to GNNs

报告人简介：

Zengfeng Huang is currently an Associate Professor in the School of Data Science, Fudan University. Before that he was a Research Fellow in CSE, UNSW and a Postdoc in MADALGO, Aarhus University. He obtained his PhD at Hong Kong University of Science and Technology in CSE and B.S. degree in Computer Science from Zhejiang University. His research interests are foundations of data science, machine learning algorithms, graph analytics, and theoretical computer science. His single-authored paper, “Near Optimal Frequent Directions for Sketching Dense and Sparse Matrices”, is the winner of ICML 2018 Best Paper Runner Up Award and 2020 World Artificial Intelligence Conference Youth Outstanding Paper Nomination Award.

个人主页：

http://www.cse.ust.hk/~huangzf/

报告摘要：

Graph neural network (GNN) is a family functions computing the embedding of each node recursively from its neighbors. In the recent few years, GNN has emerged as a major tool for graph machine learning and has found numerous applications. In this talk, I will briefly talk about several classic node embedding methods used in graph algorithms and analytics, and discuss their connection to GNNs. These embeddings are not learning-based but have non-trivial theoretical guarantees. Then I will talk about our recent work, which introduces a method inspired by graph partitioning for training GNNs.

参考文献：

[1] Shengzhong Zhang, Zengfeng Huang, Haicang Zhou, Ziang Zhou, SCE: Scalable Network Embedding from Sparsest Cut, KDD 2020.

报告嘉宾：王啸 (北京邮电大学)

报告时间：2021年3月17日 (星期三)晚上20:30 (北京时间)

报告题目：Dive into the Message Passing Mechanism of Graph Neural Networks

报告人简介：

王啸，现任北京邮电大学计算机学院助理教授，硕士生导师。研究方向为图神经网络、数据挖掘与机器学习。曾任清华大学计算机系博士后研究员，天津大学博士，美国圣路易斯华盛顿大学联合培养博士，入选2020年微软亚洲研究院铸星学者计划。在IEEE TKDE, KDD, WWW, AAAI, IJCAI等领域内国际顶级期刊和会议上发表学术论文50余篇，其中一作/通讯/共一的CCF A类论文22篇，五年内总引用2400余次，ESI高被引论文1篇，1篇入选AAAI 2017 Top 10最有影响力论文榜单，成果被主流图计算平台DGL等集成。主持国家自然科学基金和CCF-腾讯犀牛鸟科研基金。担任IEEE TKDE, KDD, AAAI, IJCAI等多个权威期刊审稿人和顶级会议的（高级）程序委员会成员。

个人主页：

https://wangxiaocs.github.io/

报告摘要：

图神经网络已成为当前深度学习领域的新浪潮，是目前学术界与工业界处理图数据的重要手段之一。本次报告将系统梳理当前图神经网络的基本特性，并探讨由此引发的一系列思考。图神经网络具备良好的结构捕捉能力，低通滤波特性等特点，但也由此带来对节点特征刻画不足、信息利用欠缺等问题。本次报告将介绍在特征刻画、高频信息利用以及对不同图神经网络的统一框架研究方面的进展，为理解图神经网络带来新的视角，并赋予图神经网络更强大与全面表达能力。

参考文献：

[1] Xiao Wang, Meiqi Zhu, Deyu Bo, Peng Cui, Chuan Shi, Jian Pei. AM-GCN: Adaptive Multi-channel Graph Convolutional Networks. ACM SIGKDD 2020.

[2] Deyu Bo, Xiao Wang, Chuan Shi, Huawei Shen. Beyond Low-frequency Information in Graph Convolutional Networks. AAAI 2021.

[3] Meiqi Zhu, Xiao Wang, Chuan Shi, Houye Ji, Peng Cui. Interpreting and Unifying Graph Neural Networks with An Optimization Framework. WWW 2021.

Panel嘉宾：何向南（中国科学技术大学）

嘉宾简介：

何向南，中国科学技术大学教授、博导。研究领域：信息检索与推荐、数据挖掘、机器学习、因果推理等，在CCF A类会议和期刊发表论文80余篇，包括20余篇SIGIR长文、20余篇 WWW和KDD长文等，谷歌学术引用9000余次，h-index 40，研究成果在多个商业公司的线上系统获得应用，取得积极效果。曾获SIGIR 2016最佳论文提名奖、WWW 2018最佳论文提名奖、SIGIR 2020最佳短文提名奖等，担任多个期刊的编委/副主编，如AI Open, Frontiers in Big Data, Hans Journal of Data Mining等。主持国家自然科学基金面上项目、重点项目，科技部重点研发计划课题等。

个人主页：

http://staff.ustc.edu.cn/~hexn/

Panel嘉宾：崔鹏 (清华大学)

嘉宾简介：

崔鹏，清华大学计算机系长聘副教授，博士生导师，2010年于清华大学获得博士学位。研究兴趣包括大数据环境下的因果推理与稳定预测、网络表征学习、社会动力学建模，及其在金融科技、智慧医疗及社交网络等场景中的应用。已在国际期刊和会议上发表论文百余篇，并先后获得7项国际会议及期刊最佳论文奖，包括中国入选数据挖掘领域顶级国际会议KDD最佳论文专刊的首篇论文。目前担任IEEE TKDE、IEEE TBD、ACM TOMM、ACM TIST等国际期刊的编委。获得国家自然科学二等奖、教育部自然科学一等奖、北京市科技进步一等奖、中国电子学会自然科学一等奖。获得中国计算机学会青年科学家奖，国际计算机协会（ACM）杰出科学家，中组部万人计划青年拔尖人才，当选为中国科协全国委员会委员、CCF YOCSEF现任主席。

个人主页：

http://pengcui.thumedialab.com/

Panel嘉宾：Jian Tang (Mila (Quebec AI institute) and HEC Montreal)

嘉宾简介：

Jian Tang is currently an assistant professor at Mila-Quebec AI Institute and HEC Montreal (business school of University of Montreal), and also a Canada CIFAR AI Research Chair. His main research interests are graph representation learning, graph neural networks, deep generative models, knowledge graphs and drug discovery. He obtained his PhD from School of EECS, Peking University in 2014, was an associate researcher at Microsoft Research Asia in 2014-2016, and was a joint postdoc fellow at University of Michigan and Carnegie Mellon University. During his PhD, he was awarded with the best paper in one of the top three machine learning conferences—ICML2014; in 2016, he was nominated for the best paper award in the top data mining conference World Wide Web (WWW); in 2020, he is awarded with Amazon and Tencent Faculty Research Award. He is one of the most representative researchers in the growing field of graph representation learning and has published a set of representative works in this field such as LINE, LargeVis, and RotatE. His work LINE on node representation learning has been widely recognized and is the most cited paper at the WWW conference between 2015 and 2019. He is the program committee and area chair of many prestige conferences in machine learning and data mining including NeurIPS, ICML, ICLR, KDD, WWW, AAAI, IJCAI, etc.

个人主页：

https://jian-tang.com/

主持人：刘昊 (宁夏大学)

主持人简介：

刘昊，宁夏大学“贺兰山学者”，副教授、硕士生导师。长期从事模式识别与计算机视觉的学术研究。主持国家自然科学基金面上项目和青年项目。发表第一作者IEEE T-PAMI长文2篇。曾获中国人工智能学会优秀博士学位论文奖、清华大学优秀博士论文奖，入选第五届中国科协青年人才托举工程。担任IEEE ICME 2020领域主席以及T-PAMI/T-IP/CVPR/ICCV/AAAI审稿人。

个人主页：

https://haoliuphd.github.io/

21-06期VALSE在线学术报告参与方式：

长按或扫描下方二维码，关注“VALSE”微信公众号 (valse_wechat)，后台回复“06期”，获取直播地址。

特别鸣谢本次Webinar主要组织者：

主办AC：刘昊 (宁夏大学)

协办AC：王文冠 (ETH Zurich)

责任AC：严骏驰 (上海交通大学)