Dr Xu is a postdoctoral researcher at National University of Singapore (NUS). He received his Ph.D. degree in computer science from the National University of Singapore in 2023. His PhD advisor is Professor Mohan S Kankanhalli. Prior to this, he obtained his B.Eng. degree in Computer Science and Technology from University of Science and Technology of China (USTC) in 2017.
His research primarily focuses on knowledge-enhanced machine learning, i.e., the integration of symbolic and formal methods with machine learning models, with the goal of better understanding the complexities in the world and building more robust, interpretable, and trustworthy models. The task he is interested in is mostly sequential relation modelling, including but not limited to visual relation detection, video understanding, and natural language processing. He is also interested in the safety and trustworthiness of powerful AI models, and how these models could affect our society and the way we live and think.
His name is pronounced as zee-way-hsu, where Xu (hsu, 徐) is his family name and Ziwei (zee-way, 子威) is his given name.
Selected Works
Ziwei Xu, Udit Sanghi, and Mohan Kankanhalli, Bullying the Machine: How Personas Increase LLM Vulnerability, 2025. [Preprint]
Ziwei Xu and Mohan Kankanhalli, Strong Preferences Affect the Robustness of Value Alignment, ICLR, 2025. [Preprint] [OpenReview] [Poster]
Ziwei Xu, Sanjay Jain, and Mohan Kankanhalli, Hallucination is Inevitable: An Innate Limitation of Large Language Models, 2024. [Preprint]
Guangzhi Wang, Yangyang Guo, Ziwei Xu, and Mohan Kankanhalli, Bilateral Adaptation for Human-Object Interaction Detection with Occlusion-Robustness, CVPR, 2024.
Ziwei Xu, Yogesh S Rawat, Yongkang Wong, Mohan S Kankanhalli, and Mubarak Shah, Don’t Pour Cereal into Coffee: Differentiable Temporal Logic for Temporal Action Segmentation, NeurIPS, 2022. (25% acceptance rate) [Project]
Ziwei Xu, Xudong Shen, Yongkang Wong, and Mohan S Kankanhalli, Unsupervised Motion Representation Learning with Capsule Autoencoders, NeurIPS, 2021. (26% acceptance rate) [Preprint] [Code] [Slides] [Video]
Ziwei Xu, Guangzhi Wang, Yongkang Wong, and Mohan S Kankanhalli, Relation-aware Compositional Zero-shot Learning for Attribute-Object Pair Recognition, IEEE Transactions on Multimedia, 2021. [Preprint] [Code]
Yaqi Xie*, Ziwei Xu*, Mohan S Kankanhalli, Kuldeep Singh Meel, and Harold Soh, Embedding Symbolic Knowledge into Deep Networks, NeurIPS, 2019. (21% acceptance rate) [Preprint]
* Equal contributions.
Hobbies and Gadgets
Astronomy Seeing/Photography
Singapore is not a perfect place for astronomy seeing.
However, the sky is quite clear when it doesn’t rain and a telescope or binoculars can be used to enjoy the night sky.
Weather forecast: cloud cover and rain in Singapore.
Planetarium: what can be seen in Singapore tonight?
Visibility plot: plot the visibility of astronomical objects.
Images on AstroBin taken in Singapore or using remote observatories.