Guanyan Ou

About Me

I am a second-year CS PhD student at Purdue University, advised by Prof. Ninghui Li. My research interests are in AI security and privacy.

Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024, Co-first Author) – Link
- Proposed a model-enhancement-based method, outperforming state-of-the-art baselines by 12.2%.
- Improved success rates of targeted black-box adversarial attacks toward computer vision classifiers.
- Employed unstructured network pruning and sharpness-aware knowledge distillation.
Patent: An Adversarial Method and Apparatus for Recognizing Misleading UI (CN116149533A) – Link
- Developed an Android app to help users identify fake UI patterns and block deceptive inputs.
- Identified real buttons via Android accessibility checks combined with computer vision results.
LLM Jailbreak (Aug 2023–May 2024)
- Induced black-box generative language models to produce harmful outputs.
- Used reinforcement learning frameworks and language models to generate natural prompts.
Diffusion Model Robustness (Sep 2024–Dec 2024)
- Using adversarial examples to protect proprietary images from being mimicked by diffusion models.

Ph.D. in Computer Science, Purdue University (Aug 2024–Expected May 2029)
B.S. in Software Engineering, Sun Yat-sen University (Sep 2020–Jun 2024)
- GPA: 3.8/4.0

Mentored Paper Review Experience (Sep 2023)
- Learned paper review processes by writing comments for 3 papers under the guidance of Prof. Weibin Wu.
RPCpico — Course Project (Jun 2022–Jul 2022)
- Implemented a Java remote procedure call (RPC) program over HTTP with Protobuf encoding; load balancing via a registry.
President, SYSU Cycling Club (Sep 2021–Sep 2022)

Programming: Python, Java, C/C++, JavaScript/TypeScript, SQL
Courses learned: Algorithm, Distributed Systems (Java), Networking (Java/C), Machine Learning, Operating System, Database