Jingyang Zhang

Research Scientist @ Virtue AI | Ph.D. @ Duke | B.Eng. @ Tsinghua

selfie_1.jpg

I’m currently a Research Scientist at Virtue AI, focusing on LLM safety (redteaming, guardrail, etc.). Previously, I got my Ph.D. from Duke ECE in 2024 where I have the honor to be advised by Prof. Yiran Chen and Prof. Hai (Helen) Li. I obtained my Bachelor degree in 2019 from the Dept. of Electronic Engineering at Tsinghua University. My dissertation research topics include adversarial machine learning and distributional shifts (specifically out-of-distribution detection). I have interned at Bosch Center for AI and Tesla over the summers.

I’m a big fan of open-source projects. Notably, I’m one of the main contributors to OpenOOD, the largest (and arguably the most well-recognized) codebase for out-of-distribution detection (on image data). I also develop lmms-finetune, a lightweight framework for fine-tuning a variety of vision LLMs on custom data, and VLM-Visualizer, an attention-based visualization that studies which part of the image the LLM looks at when predicting each token.

I (try to) write technical blogs from time to time. You can find them on Medium and 知乎.

news

Aug 04, 2025 I’m happy to join Virtue AI as a Research Scientist to work on advancing LLM safety.
Jan 22, 2025 Excited to share that our work Min-K%++ (a novel method for pretraining data detection) has been accepted to ICLR2025 as Spotlight! See [html][code][arxiv] for more details.
Jan 06, 2025 I start my job as a Machine Learning Engineer at Sciforium today. As one of the founding team members, I will be developing the latest multi-modal LLMs and touching almost every aspect of the whole spectrum, including data curation, model development, training, and deployment.
Dec 15, 2023 Excited to present our work OpenOOD v1.5 as an oral presentation at NeurIPS DistShift workshop! See [html][code][arxiv] for more details.
May 13, 2023 I’m joining Tesla as a Machine Learning Intern this summer.

selected publications

  1. NeurIPS
    DVERGE: Diversifying Vulnerabilities for Enhanced Robust Generation of Ensembles
    Huanrui Yang , Jingyang Zhang, Hongliang Dong , and 6 more authors
    In Advances in Neural Information Processing Systems , 2020
  2. WACV
    Mixture Outlier Exposure: Towards Out-of-Distribution Detection in Fine-Grained Environments
    Jingyang Zhang, Nathan Inkawhich , Randolph Linderman , and 2 more authors
    In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , Jan 2023
  3. openood_teaser.jpeg
    Openood v1.5: Enhanced benchmark for out-of-distribution detection
    Jingyang Zhang, Jingkang Yang , Pengyun Wang , and 8 more authors
    arXiv preprint, Jun 2023