Yosemite National Park, CA
2nd June 2024
Credit to Rebecca Sun

Haotian Xue 薛昊天 /haʊtiˈæn ; ʃweɪ/

Email: htxue.ai [at] gatech [dot] edu

Office: 401, MK Building; 270 Ferst Dr, Atlanta, GA 30332

I am currently a 2nd-year ML Ph.D. student at ML@GaTech, advised by Dr. Yongxin Chen. Previously, I obtained my B.E. of Computer Science from Shanghai Jiao Tong University with honor in 2022. I was a Visiting Student at MIT CSAIL supervised by Prof. Josh Tenenbaum, working closely with Yunzhu Li and Fish Tung. Feel free to contact me if you want to discuss or collaborate!

[Google Scholar] [Github] [Twitter] [Curriculum Vitae]
News 📢
  • [2024.05] I started as a research intern at Nvidia DIR Group !
  • [2024.05] We propose DP-Attacker, a framework to attack Diffusion-based policy generator.
  • [2024.04] We release PDM-Pure, a universal purifier for protection against diffusion models
  • [2024.03] I received the ICLR'2024 travel award, thanks!
  • [2024.01] Our Diff-Protect is accepted by ICLR'2024!
  • [2023.10] I was awarded the NeurIPS'2023 scholar award, thanks!
  • [2023.10] I was invited as a reviewer for TPAMI.
  • [2023.10] We propose Diff-Protect, a more effective protection framework against AI mimicry.
  • [2023.09] Diff-PGD and 3D-IntPhys are accepted by NeurIPS'2023!
  • [2023.08] I was invited as a reviewer for ICLR'2024.
  • [2023.05] We propose Diff-PGD, a diffusion-based adv-sample generation framework.
  • [2022.12] I am selected as the Top Reviewer of NeurIPS 2022.
  • [2022.10] Our Distance-Transformer is accepted to EMNLP2022 Findings.
  • [2022.08] I will start as a PhD student at (ML@GT) starting from 2022Fall.
Research Interest

My research interest lies in broad aspects of Machine Learning topics such as Computer Vision and Natural Language Processing. From 2022 to now, I focus on the following topics:

  • Generative AI & Robust AI: I am interested in building more Robust AI Systems, especially for GenAI such as Diffusion Models and LLMs. I did investigation in adversarial attacks for diffusion models [NeurIPS'23, Arxiv'24] and privacy protection against diffusion-based mimicry [ICLR'24, Arxiv'24] .
  • Compositional and Explainable AI: learning Compositional and Explainable representation, stuctures or algorithms for deep learning problems in e.g. Computer Vision and Natural Language Processing. I did research on Explanable AI [Arxiv'19, Arxiv'20] and learning compositional structure for vision tasks[NeurIPS'23, Arxiv'24] and NLP tasks [EMNLP'22] .
Research Experience
  • [2023-2024]: Collaborate with Dr. Animesh Garg, GaTech, on DM+RL
  • [2023-2024]: Collaborate with Dr. Bin Hu from UIUC and Dr. Alexandre Araujo from NYU on DM+Robustness
  • [2022-Current]: Work as a GRA Ph.D. student at FLAIR lab with Dr. Yongxin Chen
  • [2022-2023]: Work as a visiting student at MIT, advised by Prof. Josh Tenenbaum, Yunzhu Li and Fish Tung
  • [2021-2021]: Work as a research intern at NLC group, Microsoft Research, advised by Dr. Lei Cui
  • [2021-2022]: Work as a research intern at John Hopcroft Center, advised by Dr. Zhouhan Lin on NLP
  • [2020-2021]: Work as a research intern at John Hopcroft Center, advised by Dr. Quanshi Zhang on XAI
Reviewer Experience

I have reviewed 20+ papers for ML conferences like NeurIPS@22/23/24, ICLR@24 and ICML@22/23/24.

Publications ( show selected / show all by date / show all by topic )

Topics: Vision / NLP/ Robot Learning/ GenAI/ Explainable AI / (*/†: indicates equal contribution.)

RefDrop: Controllable Consistency in Image or Video Generation via Reference Feature Guidance
Jiaojiao Fan, Haotian Xue, Qinsheng, Zhang, Yongxin Chen

[Arxiv] [Project Website]

Diffusion Policy Attacker: Crafting Adversarial Attacks for Diffusion-based Policies
Yipu Chen* , Haotian Xue*, Yongxin Chen

[Arxiv] [Project Website]

Pixel is a Barrier: Diffusion Models Are More Adversarially Robust Than We Think
Haotian Xue, Yongxin Chen

[In Submission] [GitHub]

Towards More Effective Protection Against Diffusion-Based Mimicry with Score Distillation
Haotian Xue, Chumeng Liang*, Xiaoyu Wu*, Yongxin Chen

[ICLR 2024] [GitHub] [Poster]

Diffusion-Based Adversarial Sample Generation for Improved Stealthiness and Controllability
Haotian Xue, Alexandre Araujo, Bin Hu, Yongxin Chen

[NeurIPS 2023] [GitHub] [Poster]

3D-IntPhys: Towards More Generalized 3D-grounded Visual Intuitive Physics under Challenging Scenes
Haotian Xue, Antonio Torralba, Joshua B. Tenenbaum, Daniel LK Yamins, Yunzhu Li, Hsiao-Yu Tung

[NeurIPS 2023] [AIhub] [Poster]

[Abridged in CVPR 2023 3DVR & Precognition]

Syntax-guided Localized Self-attention by Constituency Syntactic Distance
Shengyuan Hou*, Jushi Kai*, Haotian Xue*, Bingyu Zhu, Bo Yuan, Longtao Huang, Xinbin Wang, Zhouhan Lin

[EMNLP2022, Findings] [GitHub]

Learning to Adaptively Incorporate External Syntax through Gated Self-Attention
[TBA]

[In Submission]

A Hypothesis For The Cognitive Difficulty of Images
Xu Chen, Xin Wang, Haotian Xue, Zhengyang Liang, Xin Jin, Quanshi Zhang

[Arxiv]

Evaluation of Attribution Explanations without Ground Truth
Hao Zhang, Haotian Xue, Jiayi Chen, Yiting Chen, Wen Shen, Quanshi Zhang

[OpenReview]

Active Adversarial Learning
Haotian Xue,

Advisor: Nanyang Ye

[Bachelor Thesis]