VALSE Webinar 20220810-20期 总第287期 图神经网络及其在结构建模中的应用 ... ...

报告嘉宾：高弘扬 (Iowa State University)

报告题目：Molecular Representation Learning via Heterogeneous Motif Graph Neural Networks

报告嘉宾：白璐 (中央财经大学)

报告题目：Transitive-Aligned Graph Neural Networks for Graph Classification

Panel嘉宾：

高弘扬 (Iowa State University)、白璐 (中央财经大学)、严骏驰 (上海交通大学)、杨旭 (西安电子科技大学)

Panel议题：

1. 图神经网络与CNN以及Transformer的关系是什么？有没有一种可能对这些模型进行统一？

2. 图神经网络在计算机视觉和机器学习中有哪些典型的应用场景？

3. 视觉和机器学习中诸多问题 (聚类、匹配等)都可以表示成图优化问题，图神经网络在优化方面的技术优势和最新进展如何？

4. 在GNN的理论研究中，有很多学者关注深度GNN的设计。那么在实际应用中，有没有必要构建深度图神经网络模型？深度GNN的优势在哪里？如何取舍深层的 GNN 网络会导致过度平滑 的问题？

5. 目前主流的GNN通常假设输入图结构是完整的正确的，对于图结构不完美的场景，GNN的这一假设会带来怎样的负面影响，有什么有效可行的解决思路？

6. 往往假设图结构是给定存在的，如何在开放场景去研究图的构建和建模，进一步解决相关问题？

*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：高弘扬 (Iowa State University)

报告时间：2022年08月10日 (星期三)晚上20:00 (北京时间)

报告题目：Molecular Representation Learning via Heterogeneous Motif Graph Neural Networks

报告人简介：

Hongyang Gao received his Ph.D. degree from Texas A&M University, College Station, Texas, in 2020. Currently, he is an Assistant Professor in the Department of Computer Science, Iowa State University, Ames, Iowa. His research interests include machine learning, deep learning, and data mining. Before his Ph.D. work, he received his M.S. from Tsinghua University in 2012 and his B.S. from Peking University in 2009.

个人主页：

https://faculty.sites.iastate.edu/hygao/

报告摘要：

We consider feature representation learning problem of molecular graphs. Graph Neural Networks have been widely used in feature representation learning of molecular graphs. However, most existing methods deal with molecular graphs individually while neglecting their connections, such as motif-level relationships. We propose a novel molecular graph representation learning method by constructing a heterogeneous motif graph to address this issue. We build a heterogeneous motif graph that contains motif nodes and molecular nodes. Each motif node corresponds to a motif extracted from molecules. Then, we propose a Heterogeneous Motif Graph Neural Network (HM-GNN) to learn feature representations for each node in the heterogeneous motif graph. Our heterogeneous motif graph also enables effective multi-task learning, especially for small molecular datasets. To address the potential efficiency issue, we propose to use an edge sampler, which can significantly reduce computational resources usage. The experimental results show that our model consistently outperforms previous state-of-the-art models. Under multi-task settings, the promising performances of our methods on combined datasets shed light on a new learning paradigm for small molecular datasets. Finally, we show that our model achieves similar performances with significantly less computational resources by using our edge sampler.

参考文献：

[1] Yu, Zhaoning, and Hongyang Gao. "Molecular Representation Learning via Heterogeneous Motif Graph Neural Networks." International Conference on Machine Learning. PMLR, 2022.

报告嘉宾：白璐 (中央财经大学)

报告时间：2022年08月10日 (星期三)晚上20:30 (北京时间)

报告题目：Transitive-Aligned Graph Neural Networks for Graph Classification

报告人简介：

白璐，中央财经大学教授，博导，国家优青获得者，国家优秀自费留学生奖获得者，百度AI华人青年学者榜单入选者。2015年1月于英国约克大学取得博士学位，师从英国皇家工程院院士Edwin R. Hancock教授。研究方向为：基于图的结构模式识别、机器学习、量子游走、金融数据分析。主持国家自然科学基金优秀青年、面上、青年项目3项。发表国际权威期刊会议论文近100篇，代表性成果包括国际顶级期刊TPAMI, TKDE, TNNLS, TCYB, PR, 以及国际顶级会议ICML, IJCAI, ICDE, ECML-PKDD, ICDM, CIKM论文近30篇。部分研究成果已应用于科大讯飞与中国电信实际业务。担任国际期刊Pattern Recognition编委，以及责任客座编辑并组织该期刊首个关于金融人工智能特刊。曾获三项国际会议最佳或优秀论文奖 (ICIAP 2015, ICPR 2018, IEEE IEEM 2019)。

个人主页：

https://www.researchgate.net/profile/Lu-Bai-33

报告摘要：

Graph-based representations are powerful tools to model complex systems that involve data lying on non-euclidean spaces and that are naturally described in terms of relations between their components, ranging from chemical compounds to point clouds and social networks. This talk will present some of our recent works of Spatially-based GCN models based on transitive aligned graph alignment strategies, that can be directly employed for graph classification problems. We show that our new models can outperform alternative state-of-the-art graph deep learning methods as well as graph kernels.

参考文献：

[1] Lu Bai, Lixin Cui*, Yuhang Jiao, Luca Rossi, Edwin R. Hancock: Learning Backtrackless Aligned-Spatial Graph Convolutional Networks for Graph Classification. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 44(2): 783 - 798, 2022.

[2] Lu Bai, Yuhang Jiao, Lixin Cui*, Luca Rossi, Yue Wang, Philip S. Yu, Edwin R. Hancock: Learning Graph Convolutional Networks based on Quantum Vertex Information Propagation. IEEE Transactions on Knowledge and Data Engineering (TKDE), Online, 2021.

[3] Lixin Cui, Lu Bai*, Xiao Bai, Yue Wang, Edwin R. Hancock: Learning Aligned Vertex Convolutional Networks for Graph Classification, IEEE Transactions on Neural Networks and Learning Systems (TNNLS), Online, 2021.

[4] Zhihong Zhang, Yangbin Zeng, Lu Bai*, Yiqun Hu, Meihong Wu, Shuai Wang, Edwin R. Hancock: Spectral Bounding: Strictly Satisfying the 1-Lipschitz Property for Generative Adversarial Networks. Pattern Recognition (PR)105: 107179, 2020.

[5] Lu Bai, Yuhang Jiao, Lixin Cui*, Edwin R. Hancock: Learning Aligned-Spatial Graph Convolutional Networks for Graph Classification. Proceedings of European Conference on Machine Learning and Knowledge Discovery in Databases (ECML-PKDD)1: 464-482, 2019.

Panel嘉宾：严骏驰 (上海交通大学)

嘉宾简介：

严骏驰，上海交通大学计算机系副教授。科技部2030新一代人工智能青年科学家项目负责人，教育部人工智能资源建设深度学习首席专家。主持国家自然基金委、上海市科委等多个项目。曾任IBM中国研究院认知计算首席研究员。获中国计算机学会优博、人工智能学会吴文俊优青等科技奖励。主要研究方向为机器学习，特别是与运筹优化、量子计算、隐私安全等领域的结合。发表CCF-A类第一/通讯作者论文80余篇，谷歌学术引用7000次。任ICML, NeurIPS, CVPR等会议的领域主席、Pattern Recognition期刊编委。

个人主页：

https://thinklab.sjtu.edu.cn/

Panel嘉宾：杨旭 (西安电子科技大学)

嘉宾简介：

杨旭， 2021年6月毕业于西安电子科技大学，现为西安电子科技大学电子工程学院副教授，硕士生导师。其研究方向为多模态数据的表示学习，主要包括自监督学习和图学习，在IEEE TPAMI, IEEE TIP, IEEE TNNLS, CVPR, NeurIPS, IJCAI, AAAI等AI相关顶级会议和期刊发表学术论文20余篇，相关研究成果获2021年吴文俊人工智能科学技术奖优秀博士论文奖。受邀担任NeurIPS, ICML, ICLR, CVPR, IJCAI, AAAI等会议程序委员会成员，以及IEEE TIP, IEEE TNNLS, IEEE TCYB等期刊审稿人。

主持人：江波 (安徽大学)

嘉宾简介：

江波，安徽大学，副教授，博士生导师。2015年获得安徽大学计算机科学与技术博士学位。CSIG 视觉大数据专委委员，CAA-模式识别与机器智能专委委员。主要从事结构模式识别、深度图学习以及视觉特征匹配等方向的研究。近年来，以第一/通讯作者在计算机领域国际顶级CCF-A会议CVPR, NeurIPS, AAAI 等和国际权威期刊 IEEE T-PAMI, IJCV, IEEE T-NNLS及IEEE T-MM等上发表论文 20余篇。主持国家自然科学基金项目，安徽省优青项目，安徽省重点研发项目等。获 ACM合肥分会新星奖，安徽省计算机学会自然科学一等奖等。担任《中国图象图形学报》青年编委，是国际权威期刊IEEE T-PAMI, IEEE T-IP, PR等审稿人。

特别鸣谢本次Webinar主要组织者：

主办AC：江波 (安徽大学)

协办AC：杨旭 (西安电子科技大学)