🧑🎓 About Me
I graduated in 2024 with a Bachelor’s degree in Artificial Intelligence from Chien-Shiung Wu College, Southeast University (SEU, 985 & 211). Afterwards, I was admitted without examination to the Master’s program at Beijing Electronic Science and Technology Institute (BESTI), where I initially considered a civil service career after encountering some early research setbacks.
However, I am fortunate to be advised by Dr. Xiaojun Jia, a postdoctoral researcher at Nanyang Technological University, and to collaborate with Weixin Wang, Haoxuan Ma, and many other brilliant peers. With Dr. Jia’s kind encouragement and recommendation, I had the privilege of completing a one-year research internship at Alibaba Security, where I was mentored by Dr. Ranjie Duan.
Building on this experience, I am honored to have been selected for the 2026 Tencent Rhino-Bird Research Elite Program. In this joint academic-industry initiative, I am currently engaged in a one-year research project co-mentored by Dr. Jolin Xia from Tencent and Prof. Xingjun Ma at Fudan University. These diverse and enriching collaborations have consistently broadened my research horizons and deepened my commitment to pursuing a Ph.D. in trustworthy AI.
📝 Selected Papers
EcoAlign: An Economically Rational Framework for Efficient LVLM Alignment [Paper]
Ruoxi Cheng, Haoxuan Ma, Teng Ma, Hongyi Zhang
The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026).
Inverse Reinforcement Learning with Dynamic Reward Scaling for LLM Alignment [Paper]
Ruoxi Cheng, Haoxuan Ma, Weixin Wang, Ranjie Duan, Jiexi Liu, Xiaoshuang Jia, Simeng Qin, Xiaochun Cao, Yang Liu, Xiaojun Jia
The Fourteenth International Conference on Learning Representations (ICLR 2026)
PBI-Attack: Prior-Guided Bimodal Interactive Black-Box Jailbreak Attack for Toxicity Maximization [Paper]
Ruoxi Cheng, Yizhong Ding, Shuirong Cao, Ranjie Duan, Xiaoshuang Jia, Shaowei Yuan, Zhiqiang Wang, Xiaojun Jia
The 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025) Main Conference
AOD: Adversarial Orthogonal Disentanglement for LVLM Hallucination Mitigation [Paper]
Ruoxi Cheng, Haoxuan Ma, Zhengfei Hai, Yiyan Huang, Ranjie Duan, Tianle Zhang, Xu Yang, Ziyi Ye, Xingjun Ma
Under review at MM 2026.
Membership Inference for Contrastive Pre-training Models with Text-only PII Queries [Paper]
Ruoxi Cheng, Yizhong Ding, Jian Zhao, Hongyi Zhang, Haoxuan Ma, Tianle Zhang, Yiyan Huang, Xuelong Li
Under review at IEEE Transactions on Information Forensics and Security (TIFS).
🎖 Honors and Awards
- 2025, First Prize (Rank 3), “Huawei Cup” National Cybersecurity Innovation Competition
- 2025, Second Prize (Top 12%), “Huawei Cup” China Graduate Mathematical Modeling Competition
- 2025, National Scholarship (Top 1%)
- 2025, Grand Prize (Rank 1), National Cybersecurity Attack and Defense Software Competition
- 2025, Second Prize (Top 5%), National Information Security Contest & “Great Wall Cup” Information Security Triathlon
- 2025, Second Prize, China’s Innovation Challenge on Artifcial Intelligence Application Scene (CICAS 2025)
- 2024 & 2025, First‑Class Academic Scholarship
- 2025, Second Prize, National Software Innovation Competition — North China Region
- 2024, Outstanding Undergraduate Thesis (Top 3%)
- 2021, Merit Student Award of Southeast University
- 2021, Second Prize, National Undergraduate Mathematics Competition
- 2018 & 2019, First Prize, Chinese Mathematical Olympiad — Jiangsu Province
📖 Educations
- 2024.09 – Present, M.S., Beijing Electronic Science and Technology Institute, Beijing, China
- 2020.09 – 2024.06, B.S., Chien-shiung Wu College, Southeast University, Nanjing, China
💻 Internships
Tencent Yuanbao (Rhino-Bird Research Elite Program) May 2026 – Present
(Industrial Supervisor: Jolin Xia | Academic Supervisor: Xingjun Ma)
- Research Project: Self-evolving Agent Systems for Complex Reasoning Tasks.
Alibaba Security Feb 2025 – Apr 2026
(Supervisor: Ranjie Duan)
-
Engaged in alignment training and evaluation of LLMs, contributing to the technical report:
- Oyster-I: Beyond Refusal — Constructive Safety Alignment for Responsible Language Models (Alibaba AAIG) [paper] — Principal Contributor (Fifth Author)
-
Co-inventor on two Alibaba Innovation Proposal patents:
- User–Model Interactive Security Guidance Mechanism Based on Game Theory — Fifth Inventor
- A Method for Constructing Chinese–English Safety Evaluation Datasets Based on Inference Complexity Grading — Sixth Inventor
💃 Skills & Interests
- Languages: English – IELTS 7.0 (Listening 7.5, Reading 8.0, Writing 6.5, Speaking 6.0); CET-6: 585
- Programming: Python, C++, PyTorch, TensorFlow, MySQL, Navicat, SPSS
- Security: NISP Level-2 Certification, administered by China Information Security Evaluation Center
- Chinese Dance: Level-10 Excellence; First Prize, Solo Dance Competition, Wuxi, Jiangsu Province (2015, 2018)
🌍 Social Service
- Sep 2024 – now, Reviewer for conferences and journals including NeurIPS, ACL and TPAMI
- Jul 2021 – Aug 2021, Village Elementary School Teaching, Xiangyang, Hubei – Volunteer, National Education Support Project
- Jun 2019 – Sep 2021, Talented Youth Initiative, Peking University – Vice Leader of Applied Science & Engineering Study Group