Shimin DI 邸世民
Research Assistant Professor
Department of Computer Science and Engineering
Hong Kong University of Science and Technology
Clear Water Bay, Kowloon, Hong Kong
- sdiaa AT connect DOT ust DOT hk
- CYT Room 3208A (Lift 19-20)
- Google Scholar DBLP
News
- Invited to give a talk "Learning from Emergence" to SAP
- Serving as the KDD'24 PC Member
- Serving as the ICML'24 PC Member
Biography
Dr. DI is a Research Assistant Professor in the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology. He was a postdoctoral researcher at HKUST working with Prof. Xiaofang ZHOU. Prior to that, he obtained his bachelor's degree in Mathematics from Hunan University, and subsequently completed his master's (2017) and doctoral degrees in Computer Science and Engineering from HKUST (2022), under the supervision of Prof. Lei CHEN. He did an internship at 4Paradigm Inc., and was mentored by Dr. Quanming YAO.
Teaching
- COMP 4211: Machine Learning (Spring 2024)
Research
The past research of Dr. DI focuses on graph representation learning, especially knowledge graph embedding and graph neural network. Heterogenous graph, as a data structure composed of multiple types of nodes and edges, is currently an important way to store and organize human knowledge and human behavior due to its ability to describe diverse correlations between multiple data objects. Heterogeneous graphs are ubiquitous in many subject areas, such as protein interaction networks in natural sciences and social networks in social sciences. In biological information modeling, heterogeneous graphs can be used to model the interactions between genes, proteins, compounds and other biomolecules, thereby helping researchers to deeply explore the structure and function of biological systems; in guiding social opinions, heterogeneous graphs can establish relationships between people, tweets/short videos, and comments, and can influence public cognition and behavior by modifying the information dissemination of social networks.
His past works mainly include graph representation model design and graph representation learning&training. Orienting to heterogeneous graphs in different scenarios, the graph representation learning model maps nodes, edges, and subgraphs from the discretized state to the continuous vector space for serving downstream tasks. Graph representation learning&training study the possible impact of learned graph representations on the different scenarios, so as to modify the graph representation to adapt to the domain-specific tasks and scenarios.
- Automated Graph Representation Model Design. (1) Techniques for Knowledge Graphs: [ICDE 2021] (improving the efficiency of model design), [WWW 2021] (saving the complexity of model design), [WWW 2023] (covering more types of knowledge graphs); (2) Techniques for Heterogeneous Graphs: [NeurIPS 2021], [KDD 2023], [ICDE 2024].
- Robust Graph Representation Learning&Training. (1) Techniques for Robust Training: [ICDE 2022], [NeurIPS 2022], [ICDE 2024], [ICDE 2024], [VLDB 2024], (2) Techniques for Incremental Learning: [SIGMOD 2023], [ICDE 2024].
Academic Service
- Program Committee Member: SIGKDD 2024/2023/2022, ICDE 2023, AAAI 2022/2021, IJCAI 2020, SDM 2024/2022, WSDM 2022
- Conference Reviewer: WWW 2024, ICML 2024/2023, NeurIPS 2023/2022, CVPR 2023
- Journal Reviewer: TKDE, TOIS