Kaijie Zhu
"To see the world, things dangerous to come to, to see behind walls, draw closer, to find each other and to feel."
zhukaijie2021@ia.ac.cn
Beijing, China
I’m a third-year Master student at the State Key Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences. I have also spent time at Microsoft, advised by Prof. Jingdong Wang and Prof. Xing Xie.
My research interest lies in the development of trustworthy AI systems and evaluation of foundation models.
- Trustworthy AI:
- Reinforce the robustness of foundation models to unexpected inputs, such as adversarial examples, jailbreak prompts, etc.
- Detecting AI-Generated Content (AIGC).
- Evaluation of foundation models:
- Dynamic evaluation for test data contamination issue.
- New evaluation measurements for generation models.
- Evaluation benchmarks reflecting diverse real-world scenarios.
Please refer to my statement of purpose for details!
news
| May 2, 2024 | DyVal 2 is accepted by ICML 2024. |
|---|---|
| Jan 17, 2024 | DyVal is accepted by ICLR 2024 as a spotlight paper! |
| Oct 22, 2023 | I am looking for a Ph.D. position in 2024 Fall! |
| Jul 18, 2023 | Our paper “Improving Generalization of Adversarial Training via Robust Critical Fine-Tuning” is accepted by ICCV 2023! |
selected publications
-
DyVal: Graph-informed Dynamic Evaluation of Large Language ModelsICLR 2024 (Spotlight, Top 5%), 2023 -
PromptBench: Towards Evaluating the Robustness of Large Language Models on Adversarial PromptsarXiv preprint arXiv:2306.04528, 2023 -
Improving Generalization of Adversarial Training via Robust Critical Fine-TuningIn ICCV 2023, 2023 -
DyVal 2: Dynamic Evaluation of Large Language Models by Meta Probing AgentsIn , 2024
