Jiacheng (James) Zhang

Jiacheng (James) Zhang

PhD candidate at Unimelb

Hi there, I am Jiacheng (James) Zhang. I am currently a visiting student at the Tsinghua Safe and Trustworthy AI Research (T-STAR) Lab at Tsinghua University, hosted by Dr. Yinpeng Dong, and a Ph.D. candidate at Trustworthy Machine Learning and Reasoning (TMLR) group in the Faculty of Engineering and Information Technology, the University of Melbourne, under the supervision of Dr. Feng Liu and Prof. Ben Rubinstein. I was a research intern in the TikTok Trust & Safety (TnS) Search Team, mentored by Dr. Haoyu He and Dr. Sen Zhang, and Previously, I was an associate member of Sea AI Lab (SAIL), mentored by Dr. Tianyu Pang and Dr. Chao Du. I received my Honours degree with a University Medal (Top 1 in my major) from the University of Sydney, where I was fortunate to learn from A/Prof. Tongliang Liu. Prior to that, I earned my Bachelor’s degree from the University of Melbourne.

I am passionate about advancing the field of trustworthy machine learning, with a long-term vision of enabling safe, reliable, and ethically aligned AI systems that can be responsibly deployed in real-world environments. My current research interests span three complementary directions, including but not limited to: (1) AI Safety Evaluation: Developing rigorous benchmarks, red-teaming methods, and statistically reliable evaluation frameworks to assess the safety of language models and AI agents; (2) AI Reliability: Improving factual consistency, multimodal grounding, and dependable behavior of AI systems under uncertainty; (3) Adversarial Robustness: Advancing robust learning and inference to defend AI systems against adversarial attacks and distribution shifts.

In an era where AI systems are increasingly embedded in high-stakes, real-world decision-making domains, I firmly believe that their foundations must be trustworthy and safe: not as an afterthought, but as a prerequisite for responsible and sustainable societal integration.

Please feel free to email me for research, collaborations, or a casual chat.

📖 Educations

  • 2023.08 - present, the Unversity of Melbourne (Unimelb), Ph.D. in Engineering and IT, advised by Dr. Feng Liu and Prof. Ben Rubinstein.
  • 2022.07 - 2023.07, the University of Sydney (USYD), B.Sc. in Data Science (Honours), advised by A/Prof. Tongliang Liu.
  • 2019.03 - 2022.03, the University of Melbourne (Unimelb), B.Sc. in Data Science.

💻 Research Experience

📝 Selected Working Papers

* Co-first author, ✉️ Corresponding author.

Preprint
sym

SafePyramid: A Hierarchical Benchmark for In-context Policy Guardrailing
Jiacheng Zhang, Haoyu He✉️, Sen Zhang✉️, Shen Wang, Xiaolei Xu, Yuhao Sun, Meng Shen, Feng Liu.
In ArXiv, 2026. [website] [paper] [huggingface] [code]

Preprint
sym

SAVAA: Mitigating Hallucinations in LVLMs via Step-wise Adaptive Visual Attention Amplification
Jiacheng Zhang, Feng Liu, Chao Du, Tianyu Pang✉️.
In ArXiv, 2026. [paper] [code]

📝 Selected Publications

* Co-first author, ✉️ Corresponding author.

NeurIPS 2026
sym

USAD: Uncertainty-aware Statistical Adversarial Detection
Zhijian Zhou*, Xunye Tian*, Jiacheng Zhang*, Zesheng Ye✉️, Yiyi Guo, Donghao Zhang, Liuhua Peng, Feng Liu✉️.
In Conference on Neural Information Processing Systems (NeurIPS 2026). [paper] [code]

TMLR 2026
sym

Semantic-aware Adversarial Fine-tuning for CLIP
Jiacheng Zhang, Jinhao Li, Hanxun Huang, Sarah M. Erfani, Benjamin I.P. Rubinstein, Feng Liu✉️.
In Transactions on Machine Learning Research (TMLR 2026). [paper] [code]

ICML 2025
sym

One Stone, Two Birds: Enhancing Adversarial Defense Through the Lens of Distributional Discrepancy
Jiacheng Zhang, Benjamin I.P. Rubinstein, Jingfeng Zhang, Feng Liu✉️.
In International Conference on Machine Learning (ICML 2025). [paper] [code] [poster]

ICML 2025
sym

Sample-specific Noise Injection for Diffusion-based Adversarial Purification
Yuhao Sun*, Jiacheng Zhang*, Zesheng Ye*, Chaowei Xiao, Feng Liu✉️.
In International Conference on Machine Learning (ICML 2025). [paper] [code] [poster]

ICML 2024
sym

Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training
Jiacheng Zhang, Feng Liu✉️, Dawei Zhou, Jingfeng Zhang, Tongliang Liu✉️.
In International Conference on Machine Learning (ICML 2024). [paper] [code] [poster]

🎖 Honours and Awards

  • 2026.03, Kingston AI Symposium Travel Scholarship.
  • 2025.07, Optiver PhD Quant Lab Travel Scholarship.
  • 2025.06, ICML 2025 Travel Award.
  • 2025.05, ICML 2025 Top Reviewer (Top 2%).
  • 2025.03, Google Student Research Travel Scholarship.
  • 2024.06, ICML 2024 Travel Award.
  • 2023.07, Amazon Best UG Senior Project Award.
  • 2023.07, University Medal of USYD (Top 1 in Major, Highest Honour for Undergraduates).
  • 2023.06, Melbourne Research Scholarship of Unimelb.
  • 2021.02, Mathematics and Statistics Vacation Research Scholarship of Unimelb.

✍️ Academic Services

  • Conference Reviewer for ICLR 2024/2025/2026, ICML 2024/2025/2026, NeurIPS 2024/2025/2026, CVPR 2026, etc.
  • Journal Reviewer for IEEE TPAMI, IEEE TIFS, TMLR, Machine Learning, NEUNET, etc.

💬 Invited Talks

  • 2025.12, Invited Talk on “How to Improve Model Robustness? A Distributional-Discrepancy Perspective” @Unimelb NLP Group Seminar, Melbourne. [slides]
  • 2025.11, Invited Talk on “How to Improve Model Robustness? A Distributional-Discrepancy Perspective” @TMLR Seminar, Melbourne. [slides]
  • 2025.08, Invited Talk on “Leveraging Distributional Discrepancies for Accuracy-robustness Trade-off” @AI TIME, Online. [slides]
  • 2025.06, Invited Talk on “Leveraging Distributional Discrepancies for Accuracy-robustness Trade-off” @AI TIME, Online. [slides] [video]
  • 2025.01, Invited Talk on “Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training” @Unimelb, Melbourne. [slides]
  • 2024.11, Invited Talk on “Improving Accuracy-robustness Trade-off via Pixel Reweighted Adversarial Training” @AJCAI 2024, Melbourne. [slides]

🧑‍🏫 Teaching Assistants

  • Teaching Assistant for SWEN20003: Object Oriented Software Development (2024 S2 / 2025 S1 / 2025 S2).
  • Teaching Assistant for COMP20008: Elements of Data Processing (2024 S1 / 2024 S2 / 2025 S1 / 2025 S2).