Graph Neural Networks Enter the Transformer Era – Seminar by Dr. Zheng Shuxin

The LOGS (Graph Learning Seminar) series periodically hosts talks on graph learning, inviting experts, researchers, and top conference authors to share insights. This session featured Dr. Zheng Shuxin from Microsoft Research.

Report Time: July 9, 2022 (Saturday), 11:00 AM Beijing Time.

Topic: Graph Neural Networks Enter the Transformer Era.

Speaker: Dr. Zheng Shuxin (Microsoft Asia Research Institute).

Host: Zhou Min (Huawei Noah’s Ark Lab).

Abstract: As a weak inductive bias model, the Transformer has become dominant in natural language processing and computer vision. Recent research shows a similar trend in graph data processing. The talk first reviews the Transformer model and its successes in CV and NLP, then presents recent breakthroughs and benefits of applying Transformers to graph data, and finally discusses future development directions.

Reference: [1] Ying C, Cai T, Luo S, et al. Do transformers really perform badly for graph representation? Advances in Neural Information Processing Systems, 2021, 34: 28877-28888.

Speaker Biography: Dr. Zheng Shuxin earned his Ph.D. jointly from the University of Science and Technology of China and Microsoft, joining Microsoft Asia Research in 2019 as a senior researcher. His work focuses on deep learning and graph neural network algorithms, with publications in top conferences (ICML, NeurIPS, ICLR, CVPR, ECCV) and high‑impact journals. Notable contributions include the Graphormer model, which won multiple AI molecular modeling competitions, and an AI‑based atmospheric modeling system adopted in China’s 14th Five‑Year Plan. He serves as a reviewer for major conferences and journals and teaches machine learning courses at Tsinghua University and Microsoft AI School.

Call to Action: Reply “logs” to the public account to obtain the live streaming link.