Skip to content
View Thecommonirin's full-sized avatar
11 followers · 20 following
  • Zhejiang University

Achievements

Achievement: Starstruck

Achievements

Achievement: Starstruck

Block or report Thecommonirin

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Thecommonirin/README.md

TheCommonIrin ✨

I am a Ph.D. candidate at the Visual Intelligence and Pattern Analysis (VIPA) Lab, Zhejiang University, co-advised by Prof. Zunlei Feng and Prof. Mingli Song. I am also deeply guided by Prof. Jian Lou throughout my research.

My research focuses on Trustworthy Generative AI (safety, alignment, and transparency) and LLM-based Agents & Decision Making. I closely collaborate with Prof. Ruoxi Jia and Prof. Tianwei Zhang.

🏠 thecommonirin.github.io

πŸ“š Activities

  • πŸŽ‰ [CCS 2026] "Understanding and Preserving Safety in Fine-Tuned LLMs" was accepted.
  • πŸŽ‰ [ACL 2026] "Token-Level Inference-Time Alignment for Vision-Language Models" was accepted to Findings of ACL.
  • πŸŽ‰ [ICLR 2026] "Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance" was accepted.
  • πŸ“ [PR 2026] "Self-Improved Holistic Alignment for Preference Enhancement" was accepted by Pattern Recognition.
  • πŸ‘¨β€πŸ’» Serving as a Reviewer for ICML, ICLR, NeurIPS, CVPR, ICCV, ECCV, TIFS, ACL, EMNLP, etc.

Pinned Loading

  1. Qresafe Qresafe Public

    Q-resafe:Assessing Safety Risks and Quantization-aware Safety Patching for Quantized Large Language Models (ICML'2025)

    Python 16 4

  2. CASPO CASPO Public

    Step-level Confidence-aware Optimization for Large Reasoning Models

    Python 2

  3. SHAPE SHAPE Public

    SHAPE: Self-Improved Holistic Alignment for Preference Enhancement (Pattern Recognition'2026)

    Python 2

  4. TITA TITA Public

    Token-level Inference-Time Alignment for Vision-Language Models (ACL'2026)

    Python 1 1

  5. SafeEnd SafeEnd Public

    Explaining many-shot jailbreaks as context-induced drift and introducing SafeEnd: a training-free defense that restores safety alignment via a single fixed demonstration.

    3

  6. Kevin-Zh-CS/QuadA Kevin-Zh-CS/QuadA Public

    Code for paper: "Activation Approximations Can Incur Safety Vulnerabilities in Aligned LLMs: Comprehensive Analysis and Defense", USENIX Security 2025.

    Python 6