Zhendong Wang

Austin, Texas, US

I am a Senior Researcher at the Microsoft GenAI Team. I completed my PhD in Statistics and Data Science at the University of Texas at Austin, where I was supervised by Prof. Mingyuan Zhou. Prior to UT Austin, I earned a Master’s degree in Data Science from Columbia University and a Bachelor’s degree from Tongji University in China. During my undergraduate studies, I spent a year as an exchange student at the University of California, Berkeley.

Research Interests

My research interests span deep generative models, reinforcement learning, and their applications. Specifically, I am interested in:

Deep Generative Models: diffusion models, GANs, and related methods.
Reinforcement Learning: online/offline RL, imitation learning, and policy optimization.
Multimodal Large Language Models: interleaved text-image generation, preference optimization algorithms, and related advancements.
Uncertainty Quantification: conformal predictions and robust modeling techniques.

I am always open to collaborations, discussions, and exploring new opportunities. Feel free to reach out if you’re interested in my research or would like to discuss potential projects. If you’re seeking intern opportunities, I’d be happy to connect!

news

Dec 02, 2024	I am joining Microsoft as a Senior Researcher.
Nov 01, 2024	Our paper One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation was public at Nvidia Website. The work shows the broad potential of diffusion distillation for robotics.
Nov 01, 2024	We share a series of diffusion distillation works: Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation [ICML 2024] A fundamental distillation technique SiD through Fisher Divergence. One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation Coupling with CFG, SiD works well in text-to-image one-step generation. Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step Introduce the data dependency to eliminate pretraining bias and further boost the performance of SiD.
Oct 01, 2024	Our paper Diffusion Policies creating a Trust Region for Offline Reinforcement Learning was published at NeurIPS 2024 and the code was released on Github.
Mar 15, 2024	I will join NVIDIA Deep Imagination Research group led by Ming-Yu Liu for 2024 summer internship.
Jan 01, 2024	Our new paper Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Prompts is now public on ArXiv and the code has been publicly released on Github. Continue the Part-time Internship with Microsoft GenAI team for Fall 2023 and Spring 2024.
Jan 01, 2024	Our paper In-Context Learning Unlocked for Diffusion Models has been accepted by NeurIPS 2023 and the code has been publicly released on Github with diffusers supported. Our paper Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models has been accepted by NeurIPS 2023 and the code has been publicly released on Github.
May 01, 2023	Our new paper In-Context Learning Unlocked for Diffusion Models was public on arXiv and the code was publicly released on Github. Our new paper Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models was public on arXiv and the code will be released soon. I am joining Microsoft Azure AI team for 2023 summer internship.
Jan 30, 2023	Our new paper Diffusion Policies as an Expressive Policy Class for Offline Reinforcement Learning was accepted by ICLR 2023 and the code was publicly released on Github. Our new paper Diffusion-GAN: Training GANs with Diffusion was accepted by ICLR 2023 and the code was publicly released on Github. Our new paper Probabilistic Conformal Prediction Using Conditional Random Samples was accepted by AISTATS 2023 and the code was publicly released on Github.
Sep 15, 2022	Joined Microsoft Azure AI team for my part-time internship for Fall 2022 and Spring 2023.

selected publications

Preprint

One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation

Zhendong Wang, Zhaoshuo Li, Ajay Mandlekar, and 5 more authors

ArXiv Preprint, 2024

Bib PDF Code

@article{wang2024onedp,
  title = {One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation},
  author = {Wang, Zhendong and Li, Zhaoshuo and Mandlekar, Ajay and Xu, Zhenjia and Fan, Jiaojiao and Narang, Yashraj and Fan, Linxi and et al},
  journal = {ArXiv Preprint},
  year = {2024},
  url = {https://arxiv.org/abs/2410.21257},
}

Preprint

Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step

Mingyuan Zhou, Huangjie Zheng, Yi Gu, and 2 more authors

ArXiv Preprint, 2024

Bib PDF Code

@article{zhou2024sida,
  title = {Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step},
  author = {Zhou, Mingyuan and Zheng, Huangjie and Gu, Yi and Wang, Zhendong and Huang, Hai},
  journal = {ArXiv Preprint},
  year = {2024},
  url = {https://arxiv.org/abs/2410.14919},
}

Preprint

Long and Short Guidance in Score identity Distillation for One-Step Text-to-Image Generation

Mingyuan Zhou, Zhendong Wang, Huangjie Zheng, and 1 more author

ArXiv Preprint, 2024

Bib PDF Code

@article{zhou2024sidlsg,
  title = {Long and Short Guidance in Score identity Distillation for One-Step Text-to-Image Generation},
  author = {Zhou, Mingyuan and Wang, Zhendong and Zheng, Huangjie and Huang, Hai},
  journal = {ArXiv Preprint},
  year = {2024},
  url = {https://arxiv.org/abs/2406.01561},
}

ICML 2024

Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation

Mingyuan Zhou, Huangjie Zheng, Zhendong Wang, and 1 more author

International Conference on Machine Learning 2024, 2024

Bib PDF Code

@article{zhou2024sid,
  title = {Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation},
  author = {Zhou, Mingyuan and Zheng, Huangjie and Wang, Zhendong and Huang, Hai},
  journal = {International Conference on Machine Learning 2024},
  year = {2024},
  url = {https://arxiv.org/abs/2404.04057},
}

Preprint

Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Prompts

Yueqin Yin, Zhendong Wang, Yi Gu, and 3 more authors

ArXiv Preprint:2402.10958, 2024

Bib PDF Code

@article{yin2024relative,
  title = {Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Prompts},
  author = {Yin, Yueqin and Wang, Zhendong and Gu, Yi and Huang, Hai and Chen, Weizhu and Zhou, Mingyuan},
  journal = {ArXiv Preprint:2402.10958},
  year = {2024},
  url = {https://arxiv.org/abs/2402.10958},
}

NeurIPS 2023

In-Context Learning Unlocked for Diffusion Models

Zhendong Wang, Yifan Jiang, Yadong Lu, and 5 more authors

Advances in Neural Information Processing Systems, 2023

Bib PDF Code

@article{wang2023context,
  title = {In-Context Learning Unlocked for Diffusion Models},
  author = {Wang, Zhendong and Jiang, Yifan and Lu, Yadong and Shen, Yelong and He, Pengcheng and Chen, Weizhu and Wang, Zhangyang and Zhou, Mingyuan},
  journal = {Advances in Neural Information Processing Systems},
  year = {2023},
  url = {https://arxiv.org/abs/2305.01115},
}

NeurIPS 2023

Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models

Zhendong Wang, Yifan Jiang, Huangjie Zheng, and 5 more authors

Advances in Neural Information Processing Systems, 2023

Bib PDF Code

@article{wang2023patch,
  title = {Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models},
  author = {Wang, Zhendong and Jiang, Yifan and Zheng, Huangjie and Wang, Peihao and He, Pengcheng and Wang, Zhangyang and Chen, Weizhu and Zhou, Mingyuan},
  journal = {Advances in Neural Information Processing Systems},
  year = {2023},
  url = {https://arxiv.org/abs/2304.12526},
}

ICLR 2023

Diffusion Policies as an Expressive Policy Class for Offline Reinforcement Learning

Zhendong Wang, J Jonathan Hunt, and Mingyuan Zhou

International Conference on Learning Representations, 2023

Bib PDF Code

@article{wang2023diffusioninrl,
  title = {Diffusion Policies as an Expressive Policy Class for Offline Reinforcement Learning},
  author = {Wang, Zhendong and Hunt, J Jonathan and Zhou, Mingyuan},
  journal = {International Conference on Learning Representations},
  year = {2023},
  url = {https://arxiv.org/abs/2208.06193},
}

ICLR 2023

Diffusion-GAN: Training GANs with Diffusion

Zhendong Wang, Huangjie Zheng, Pengcheng He, and 2 more authors

International Conference on Learning Representations, 2023

Bib PDF Code

@article{wang2023diffusiongan,
  title = {Diffusion-GAN: Training GANs with Diffusion},
  author = {Wang, Zhendong and Zheng, Huangjie and He, Pengcheng and Chen, Weizhu and Zhou, Mingyuan},
  journal = {International Conference on Learning Representations},
  year = {2023},
  url = {https://arxiv.org/abs/2206.02262},
}