About Me
Expected Graduation: June 2027. Open to full-time Research Scientist / Research Engineer / Member of Technical Staff roles.
Please feel free to contact me at to discuss potential opportunities.
I am a Ph.D. candidate at the Provable Responsible AI and Data Analytics (PRADA) Lab at the King Abdullah University of Science and Technology (KAUST), advised by Prof. Di Wang. I am also doing a research internship at the Microsoft Research Asia (MSRA), working with Dr. Xingxing Zhang.
Previously, I was an algorithm engineer in the Trustworthy AI Research Group at JD Explore Academy, JD.com, Inc.. I received an MPhil in Engineering and IT from The University of Sydney, advised by Prof. Dacheng Tao, and a B.Sc. in Mathematics and Applied Mathematics from the South China University of Technology, advised by Prof. Chuhua Xian.
Contact
shaopeng.fu@kaust.edu.sa
shaopengfu15@gmail.com
Research Summary
I develop principled methods and scalable infrastructure for reliable and efficient LLM post-training, adversarial robustness, scalable evaluation, and adversarial-training-inspired data synthesis. My research combines deep learning theory with system-level optimization to improve the training and evaluation of reasoning models. Recent topics include:
- RL post-training for coding LLMs and agents: Preprint’26.
- LLM adversarial training and jailbreak robustness: Preprint’26a, Preprint’26b, ICLR’26, NeurIPS’25, ICLR’24.
- Model and data privacy: Preprint’24, ICLR’22a, ICLR’22b.
If you are interested in collaborating with me or discussing my research, please feel free to contact me through email.
News
- 08/2026: We released our new paper Dual-Adversarial Safety Alignment: Cultivating Intrinsic Threat Comprehension in LRMs.
- 04/2026: Two papers on robust overfitting and private feature learning were accepted to ICML 2026. Thanks to my great collaborators!
- 04/2026: We released our new paper RefineRL: Advancing Competitive Programming with Self-Refinement Reinforcement Learning.
- 03/2026: We released our new paper Accelerating Suffix Jailbreak attacks with Prefix-Shared KV-cache.
- 01/2026: Our paper on LLM continuous adversarial training theory was accepted to ICLR 2026!
- 11/2025: I passed my Ph.D. Proposal Defense and officially became a Ph.D. candidate. Thanks to everyone who helped me during this journey!
- 09/2025: Our paper on LLM adversarial training theory was accepted to NeurIPS 2025!
Selected Publications [Full List] [Google Scholar]
* indicates co-first authors.
LLM Code Generation
- RefineRL: Advancing Competitive Programming with Self-Refinement Reinforcement Learning
[arXiv]
Shaopeng Fu, Xingxing Zhang, Li Dong, Di Wang, and Furu Wei
arXiv preprint 2026
Adversarial Robustness
Dual-Adversarial Safety Alignment: Cultivating Intrinsic Threat Comprehension in LRMs
[arXiv] [Code]
Hongli Shen*, Shaopeng Fu*, Qinbo Zhang, Jian Li, and Di Wang
arXiv preprint 2026Accelerating Suffix Jailbreak attacks with Prefix-Shared KV-cache
[arXiv] [Code]
Xinhai Wang*, Shaopeng Fu*, Shu Yang, Liangyu Wang, Tianhang Zheng, and Di Wang
arXiv preprint 2026Understanding and Improving Continuous LLM Adversarial Training via In-context Learning Theory
[Link] [arXiv] [Code]
Shaopeng Fu and Di Wang
ICLR 2026Short-length Adversarial Training Helps LLMs Defend Long-length Jailbreak Attacks: Theoretical and Empirical Evidence
[Link] [arXiv] [Video] [Code]
Shaopeng Fu, Liang Ding, Jingfeng Zhang, and Di Wang
NeurIPS 2025Theoretical Analysis of Robust Overfitting for Wide DNNs: An NTK Approach
[OpenReview] [IEEE] [arXiv] [Video] [Code]
Shaopeng Fu and Di Wang
ICLR 2024
IEEE Transactions on Information Theory
Data/Model Privacy
Pre-trained Encoder Inference: Revealing Upstream Encoders In Downstream Machine Learning Services
[arXiv] [Code]
Shaopeng Fu, Xuexue Sun, Ke Qing, Tianhang Zheng, and Di Wang
arXiv preprint 2024Robust Unlearnable Examples: Protecting Data Against Adversarial Learning
[Link] [arXiv] [Video] [Code]
Shaopeng Fu, Fengxiang He, Yang Liu, Li Shen, and Dacheng Tao
ICLR 2022Knowledge Removal in Sampling-based Bayesian Inference
[Link] [arXiv] [Video] [Code]
Shaopeng Fu*, Fengxiang He*, and Dacheng Tao
ICLR 2022
Services
- Conference Reviewer: ICML (2022-2026) / ICLR (2022-2026) / NeurIPS (2021-2026) / AISTATS (2021, 2024-2026)
- Conference Committee: CCS 2024 (Artifact Evaluation) / AAAI 2025
- Journal Reviewer: Neurocomputing / TMLR / IEEE TIT / IEEE TPAMI / IEEE TNNLS / IEEE TCYB / Springer NPL
Selected Awards
- International Collegiate Programming Contest (ICPC)
- Gold Medal: Asia Regional Contest Shenyang Site (2018; Rank: 6/186)
- Silver Medals (3x): Asia-East Continent Final Xi’an Site (2018), Asia Regional Contest Qingdao Site (2017), and Asia Regional Contest Xi’an Site (2017)
- National Scholarship (2x): 2017 & 2018
- ICML Silver Reviewer: 2026