Xun Xian

I am Xun Xian, a fourth-year Ph.D. student at University of Minnesota, supervised by Prof. Jie Ding and Prof. Mingyi Hong.

I am dedicated to investigating the trustworthiness and safety of LLMs and Diffusion models. Presently, my efforts are concentrated on developing watermarking techniques tailored for generative models, encompassing both diffusion and large language models, to ensure the safe use of generative AI.

Contact: By Email xian0044@umn.edu

news

Jan 16, 2024	Our paper titled ‘Demystifying Poisoning Backdoor Attacks from a Statistical Perspective’ has been accepted at ICLR 2024. In this paper, we developed a fundamental understanding of backdoor attacks for generative models, including both diffusion models and large language models.
Oct 24, 2023	Glad to receive the NeurIPS 2023 Scholar Award!
Sep 30, 2023	Our paper titled ‘A Unified Detection Framework for Inference-Stage Backdoor Defenses’ has been accepted at NeurIPS 2023. In this paper, we developed a unified framework for defending against backdoor attacks in both computer vision (CV) and natural language processing (NLP) applications, achieving up to a 300% improvement in terms of detection performance.
May 30, 2023	A paper is accepted at ICML 2023.
Dec 30, 2022	Glad to receive the Cisco Graduate Research Award!

selected publications

NeurIPS’23

A Unified Detection Framework for Inference-Stage Backdoor Defenses

Xun Xian, Ganghua Wang, Jayanth Srinivasa, and 4 more authors

In Thirty-seventh Conference on Neural Information Processing Systems, 2023

Bib HTML

@inproceedings{xian2023defense,
  title = {A Unified Detection Framework for Inference-Stage Backdoor Defenses},
  author = {Xian, Xun and Wang, Ganghua and Srinivasa, Jayanth and Kundu, Ashish and Bi, Xuan and Hong, Mingyi and Ding, Jie},
  booktitle = {Thirty-seventh Conference on Neural Information Processing Systems},
  year = {2023}
}

ICML’23

Understanding Backdoor Attacks through the Adaptability Hypothesis

Xun Xian, Ganghua Wang, Jayanth Srinivasa, and 4 more authors

In International Conference on Machine Learning, 2023

Bib HTML

@inproceedings{xian2023understanding,
  title = {Understanding Backdoor Attacks through the {Adaptability Hypothesis}},
  author = {Xian, Xun and Wang, Ganghua and Srinivasa, Jayanth and Kundu, Ashish and Bi, Xuan and Hong, Mingyi and Ding, Jie},
  corresponding = {Xun Xian and Ganghua Wang},
  equalcontribution = {true},
  booktitle = {International Conference on Machine Learning},
  year = {2023}
}

NeurIPS’20

Assisted learning: A framework for multi-organization learning

Xun Xian, Xinran Wang, Jie Ding, and 1 more author

In Thirty-fourth Conference on Neural Information Processing Systems, 2020

Bib HTML

@inproceedings{xian2020assisted,
  title = {Assisted learning: A framework for multi-organization learning},
  author = {Xian, Xun and Wang, Xinran and Ding, Jie and Ghanadan, Reza},
  booktitle = {Thirty-fourth Conference on Neural Information Processing Systems},
  year = {2020}
}

ICLR’24

Demystifying Poisoning Backdoor Attacks from a Statistical Perspective

Ganghua Wang, Xun Xian, Jayanth Srinivasa, and 4 more authors

In The International Conference on Learning Representations, 2024

Bib HTML

@inproceedings{xw_2024,
  title = {Demystifying Poisoning Backdoor Attacks from a Statistical Perspective},
  author = {Wang, Ganghua and Xian, Xun and Srinivasa, Jayanth and Kundu, Ashish and Bi, Xuan and Hong, Mingyi and Ding, Jie},
  corresponding = {Ganghua Wang and Xun Xian},
  equalcontribution = {true},
  booktitle = {The International Conference on Learning Representations},
  year = {2024}
}