Zihao Zhu

me.jpeg

zihaozhu@link.cuhk.edu.cn

Hi, this is Zihao Zhu (朱梓豪). I am currently a Ph.D. student in Data Science at The Chinese University of Hong Kong, Shenzhen, under the supervision of Prof. Baoyuan Wu. Previously, I received my Master’s degree from the Institute of Information Engineering at the University of Chinese Academy of Sciences in 2021.

My research interests lie in the broad area of AI security, with focuses on three main directions:

  • Security of Large Language Models: I work on understanding and addressing security challenges in LLMs, including jailbreak attacks and AI alignment. This research aims to make language models more robust while maintaining their utility.
  • Data Security in AI Systems: Data is the fuel of AI. I investigate various aspects of data security in Data-centric AI (DCAI), with particular emphasis on backdoor attacks and data quality assessment.
  • Security in Embodied AI: I explore security concerns in embodied AI systems, focusing on risk assessment for AI agents. This emerging area is crucial as AI systems become more integrated into physical environments.

If you share similar interests, please feel free to reach out. I am happy to chat and open to exploring opportunities for collaboration.

news

Feb 19, 2025 Our new paper “BoT: Breaking Long Thought Processes of o1-like Large Language Models through Backdoor Attack” is available on arXiv. Check out the code on GitHub.
Dec 07, 2024 One new preprint is available: “HMGIE: Hierarchical and Multi-Grained Inconsistency Evaluation for Vision-Language Data Cleansing” :page_facing_up:
Jan 20, 2024 Our paper “Learning to Optimize Permutation Flow Shop Scheduling via Graph-based Imitation Learning” has been accepted to AAAI 2024! :rocket:
Jan 20, 2024 Our paper “VDC: Versatile Data Cleanser based on Visual-Linguistic Inconsistency by Multimodal Large Language Models” has been accepted to ICLR 2024! :tada:

selected publications

  1. BoT: Breaking Long Thought Processes of o1-like Large Language Models through Backdoor Attack
    Zihao Zhu, Hongbao Zhang, Mingda Zhang, Ruotong Wang , Guanzong Wu, Xu Ke, and Baoyuan Wu
    arXiv preprint, 2025
  2. HMGIE: Hierarchical and Multi-Grained Inconsistency Evaluation for Vision-Language Data Cleansing
    Zihao Zhu, Hongbao Zhang , Guanzong Wu, Siwei Lyu, and Baoyuan Wu
    arXiv preprint, 2024
  3. Reliable Poisoned Sample Detection against Backdoor Attacks Enhanced by Sharpness Aware Minimization
    Mingda Zhang, Mingli Zhu, Zihao Zhu, and Baoyuan Wu
    arXiv preprint, 2024
  4. EARBench: Towards Evaluating Physical Risk Awareness for Task Planning of Foundation Model-based Embodied AI Agents
    Zihao Zhu , Bingzhe Wu, Zhengyou Zhang, Lei Han, Qingshan Liu, and Baoyuan Wu
    arXiv preprint, 2024
  5. ICLR
    vdc.png
    VDC: Versatile Data Cleanser based on Visual-Linguistic Inconsistency by Multimodal Large Language Models
    Zihao Zhu, Mingda Zhang, Shaokui Wei , Bingzhe Wu, and Baoyuan Wu
    In International Conference on Learning Representations, 2024
  6. BlackboxBench: A Comprehensive Benchmark of Black-box Adversarial Attacks
    Meixi Zheng, Xuanchen Yan, Zihao Zhu, Hongrui Chen, and Baoyuan Wu
    arXiv preprint, 2023
  7. Defenses in adversarial machine learning: A survey
    Baoyuan Wu, Shaokui Wei, Mingli Zhu, Meixi Zheng, Zihao Zhu, Mingda Zhang, Hongrui Chen, Danni Yuan, Li Liu, and Qingshan Liu
    arXiv preprint, 2023
  8. Boosting backdoor attack with a learnable poisoning sample selection strategy
    Zihao Zhu, Mingda Zhang, Shaokui Wei, Li Shen, Yanbo Fan, and Baoyuan Wu
    arXiv preprint, 2023
  9. Attacks in adversarial machine learning: A systematic survey from the life-cycle perspective
    Baoyuan WuZihao Zhu, Li Liu, Qingshan Liu, Zhaofeng He, and Siwei Lyu
    arXiv preprint, 2023
  10. NeurIPS
    backdoorbench.png
    BackdoorBench: A Comprehensive Benchmark of Backdoor Learning
    Baoyuan Wu, Hongrui Chen, Mingda Zhang, Zihao Zhu, Shaokui Wei, Danni Yuan, and Hongyuan Zha
    In Advances in Neural Information Processing Systems, 2022
  11. ICASSP
    shallow.png
    From Shallow to Deep: Compositional Reasoning over Graphs for Visual Question Answering
    Zihao Zhu
    In IEEE International Conference on Acoustics, Speech and Signal Processing, 2022
  12. PR
    gruc.png
    Cross-Modal Knowledge Reasoning for Knowledge-based Visual Question Answering
    Jing YuZihao Zhu, Yujing Wang, Weifeng Zhang, Yue Hu, and Jianlong Tan
    Pattern Recognition, 2020
  13. IJCAI
    mucko.png
    Mucko: Multi-Layer Cross-Modal Knowledge Reasoning for Fact-based Visual Question Answering
    Zihao ZhuJing Yu, Yujing Wang, Yajing Sun, Yue Hu , and Qi Wu
    In Proceedings of the International Joint Conference on Artificial Intelligence, 2020