Li Xiao (李潇)

Ph.D. student at the Department of Computer Science and Technology, Tsinghua University.

xiaoli.cst@gmail.com
Google Scholar

I am open to discussions and collaborations if you’re interested in my work!

Affiliation:
Department of Computer Science and Technology,
State Key Laboratory of Intelligent Technology and Systems (CSAI),
Tsinghua University, Beijing 100084, China;
Tsinghua Laboratory of Brain and Intelligence (THBI);

I am a final-year Ph.D. student at the Department of Computer Science and Technology in Tsinghua University, advised by Prof. Xiaolin Hu and Prof. Bo Zhang. I am a member of TSAIL Group, which is directed by Prof. Bo Zhang and Prof. Jun Zhu. I received my Bachelor’s degree at Department of Computer Science and Technology from Tsinghua University in 2020.

My research interests lie at the intersection of computer vision (CV), natural language processsing (NLP), and machine learning (ML), with the goal of developing trustworthy multimodal systems that can achieve human‐level visual understanding and out-of-distribution (OOD) generalization. I have explored the following areas:

Classical CV applications: Improving the accuracy and robustness of general object detection, instance segmentation, and other tasks.

Visual-language multimodal learning: Exploring potential issues and solutions for aligning multimodal models on OOD samples.

Diffusion models: Enhancing the OOD generalization of visual models by leveraging the ability of diffusion models.

Large language models: Investigating potential security risks of large language models, establishing efficient jailbreak attacks and defense methods for large language models.

Trustworthy machine learning: Exploring frontier problems such as federated learning, adversarial ML, and ML privacy.

Previously, I interned at 01.ai. I am currently working at Bytedance Seed, focusing on the development of cutting-edge multimodal and reasoning foundation models.

news

Aug 30, 2024	We propose the Faster-GCG algothrim, a foundamental and efficient discrete optimization approach for jailbreak attacks against large language models. Read more
Aug 1, 2024	We propose the ADBM model, which can significantly improves the robustness of visual models on OOD examples. We show theoretically and empirically that ADBM outperforms the original DDPM. This work has been accepted in ICLR 2025. Read more
Jul 7, 2024	We released the PartImageNet++ dataset and further improved the part-based recogntion models. The paper has been accepted by ECCV 2024. Read more
Mar 18, 2024	One paper on the relation between adversarial robustness and privacy is accepted by IEEE TIFS 2024. Read more
Feb 26, 2024	One paper on achiving zero-shot adversarial robustness with multimodal CLIP models is accepted by CVPR 2024. The proposed LAAT method uses language-driven anchors to guide adversarial training of vision models. Read more
May 27, 2023	One paper on how to improving robustness of object detectors with upstream adversarial pre-training is available on arXiv. Read more
Jan 18, 2023	One paper inspired by cognitive psychology theory is accepted by IEEE TPAMI 2023. The proposed ROCK method can significantly improve both adversarial robusntess and generalization on out-of-distribution examples. Read more
Jul 14, 2022	One paper is accepted by IJCV 2022, which is an extended version of the BPR paper. Read more
Mar 1, 2021	One paper on instance segmentatiion is accepted to CVPR 2021. The proposed BPR method reached 1st place on the Cityscapes leaderboard (instance segmentation track). Read more

selected publications

2025

ICLR

ADBM: Adversarial diffusion bridge model for reliable adversarial purification

Xiao Li, Wenxuan Sun, Huanran Chen, Qiongxiu Li, Yining Liu, Yingzhe He, Jie Shi, and Xiaolin Hu

International Conference on Learning Representations (ICLR), 2025

Bib arXiv Code

@article{li2024adbm,
  link = {https://openreview.net/forum?id=g0rnZeBguq},
  title = {ADBM: Adversarial diffusion bridge model for reliable adversarial purification},
  author = {Li, Xiao and Sun, Wenxuan and Chen, Huanran and Li, Qiongxiu and Liu, Yining and He, Yingzhe and Shi, Jie and Hu, Xiaolin},
  journal = {International Conference on Learning Representations (ICLR)},
  year = {2025}
}

2024

arXiv

Faster-GCG: Efficient Discrete Optimization Jailbreak Attacks against Aligned Large Language Models

Xiao Li, Zhuhong Li, Qiongxiu Li, Bingze Lee, Jinghao Cui, and Xiaolin Hu

arXiv preprint arXiv:2410.15362, 2024

Bib arXiv

@article{fastergcg,
  title = {Faster-GCG: Efficient Discrete Optimization Jailbreak Attacks against Aligned Large Language Models},
  author = {Li, Xiao and Li, Zhuhong and Li, Qiongxiu and Lee, Bingze and Cui, Jinghao and Hu, Xiaolin},
  journal = {arXiv preprint arXiv:2410.15362},
  year = {2024}
}

ECCV

PartImageNet++ Dataset: Scaling up Part-based Models for Robust Recognition

Xiao Li, Yining Liu, Na Dong, Sitian Qin, and Xiaolin Hu

European Conference on Computer Vision (ECCV), 2024

Bib arXiv Code

@article{lipartimagenetpp,
  title = {PartImageNet++ Dataset: Scaling up Part-based Models for Robust Recognition},
  author = {Li, Xiao and Liu, Yining and Dong, Na and Qin, Sitian and Hu, Xiaolin},
  journal = {European Conference on Computer Vision (ECCV)},
  year = {2024},
  publisher = {Springer}
}

CVPR

Language-Driven Anchors for Zero-Shot Adversarial Robustness

Xiao Li, Wei Zhang, Yining Liu, Zhanhao Hu, Bo Zhang, and Xiaolin Hu

In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

Bib arXiv Code

@inproceedings{li2023languagedriven,
  author = {Li, Xiao and Zhang, Wei and Liu, Yining and Hu, Zhanhao and Zhang, Bo and Hu, Xiaolin},
  title = {Language-Driven Anchors for Zero-Shot Adversarial Robustness},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages = {24686--24695},
  year = {2024}
}

2023

TPAMI

Recognizing Object by Components With Human Prior Knowledge Enhances Adversarial Robustness of Deep Neural Networks

Xiao Li, Ziqi Wang, Bo Zhang, Fuchun Sun, and Xiaolin Hu

IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2023

Bib arXiv Code

@article{rock,
  author = {Li, Xiao and Wang, Ziqi and Zhang, Bo and Sun, Fuchun and Hu, Xiaolin},
  title = {Recognizing Object by Components With Human Prior Knowledge Enhances
                    Adversarial Robustness of Deep Neural Networks},
  journal = {{IEEE} Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
  volume = {45},
  number = {7},
  pages = {8861--8873},
  year = {2023}
}