🔬 Research
This page collects my research experience and related work in more detail. The homepage highlights selected items, while this page serves as the longer-term archive for current and future research.
Jailbreak Attacks on Large Language Models
Role: Project Leader
Period: Oct 2024 - Sep 2025
Project Type: University-Level Undergraduate Innovation Project
This project studies security risks and adversarial vulnerabilities in large language models. I conducted a focused literature review on jailbreak attacks and defense strategies, then evaluated mainstream models under a range of prompt-based attack settings.
To improve attack efficiency and transferability, I further designed a reinforcement-learning-based prompt optimization framework for generating stronger adversarial prompts.
Current outcomes: Survey manuscript; one invention patent.
Scenario-Controlled Image Generation for Animal Recognition
Role: Core Team Member
Period: Oct 2024 - Sep 2025
Project Type: National-Level Undergraduate Innovation Project
This project explores controllable image generation methods and how synthetic data can improve animal recognition performance. I participated in the design of generation pipelines under scenario constraints and helped evaluate their usefulness for downstream recognition tasks.
