Chengzhi Mao

Assistant Professor, Department of Computer Science, Rutgers University
chengzhi.mao@rutgers.edu Google Scholar
Chengzhi Mao headshot

Brief Bio

I am an Assistant Professor in the Department of Computer Science at Rutgers University. Previously, I have been a Research Scientist at Google, working on Gemini and GenAI related research. I have been a Core Faculty Member at MILA, and an Assistant Professor at McGill. I completed my Ph.D. at Columbia University in 2023, advised by Carl Vondrick and Junfeng Yang. I got my BS at Tsinghua University in 2018, advised by Yuan Shen. I was also a visiting student at MIT, advised by Dina Katabi.

My research goal is to build reliable multimodal reasoning models that can serve as the “brain” of robots and computer agents. I aim to develop AI systems that integrate vision, language, action, and world knowledge to reason, plan, and act safely and smoothly in open-world environments. More broadly, my work seeks to advance trustworthy foundation models that can understand complex multimodal contexts, make robust decisions, and collaborate effectively with humans in the physical world.

I am actively recruiting highly self-motivated Ph.D. students, research interns, and postdocs. If you’re interested, please email me with your CV. Interest and experience in robotics, multimodal reasoning, or generative models will be a plus.

News

Students

Yang Li — PhD Student Yuan Qing — PhD Student Yibin Wang — PhD Student Zirui Zhang — PhD Student Shilong Xiang — Research Assistant

Selected Papers

2026
SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation thumbnail

SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation

Shilong Xiang, Zirui Zhang, Lijun Yu, Chengzhi Mao
arXiv — paper / project page
StochasT : Learning with Stochastic Turn Depth for Visual Instruction Tuning thumbnail

StochasT : Learning with Stochastic Turn Depth for Visual Instruction Tuning

Yuan Qing, Chengzhi Mao, Boqing Gong
ECCV 2026 — paper
APPLV: Adaptive Planner Parameter Learning from Vision-Language-Action Model thumbnail

APPLV: Adaptive Planner Parameter Learning from Vision-Language-Action Model

Yuanjie Lu, Beichen Wang, Zhengqi Wu, Yang Li, Xiaomin Lin, Chengzhi Mao, Xuesu Xiao
IROS 2026 — paper
S2COPE: Self-Supervised Concept Discovery via Preference Learning thumbnail

S2COPE: Self-Supervised Concept Discovery via Preference Learning

Shilong Xiang, Zirui Zhang, Chengzhi Mao
arXiv — paper / project page
Differences That Matter: Auditing Models for Capability Gap Discovery and Rectification thumbnail

Differences That Matter: Auditing Models for Capability Gap Discovery and Rectification

Qihao Liu, Chengzhi Mao, Yaojie Liu, Alan Yuille, Wen-Sheng Chu
CVPR'26 Highlight — paper / project page
LACE: Lattice Attention for Cross-Thread Exploration thumbnail

LACE: Lattice Attention for Cross-Thread Exploration

Yang Li, Zirui Zhang, Yang Liu, Chengzhi Mao
arXiv 2026 — paper
R-C2: Cycle-Consistent Reinforcement Learning Improves Multimodal Reasoning thumbnail

R-C2: Cycle-Consistent Reinforcement Learning Improves Multimodal Reasoning

Zirui Zhang, Haoyu Dong, Kexin Pei, Chengzhi Mao
CVPR 2026 — paper / code / project page
Mull-Tokens: Modality-Agnostic Latent Thinking thumbnail

Mull-Tokens: Modality-Agnostic Latent Thinking

Arijit Ray, Ahmed Abdelkader, Chengzhi Mao, Bryan Plummer, Kate Saenko, Ranjay Krishna, Leonidas Guibas, Wen-Sheng Chu
CVPR 2026 Findings — paper / code / project page
Language Instructed Vision Embeddings for Controllable and Generalizable Perception thumbnail

Language Instructed Vision Embeddings for Controllable and Generalizable Perception

Chengzhi Mao, Xudong Lin, Wen-Sheng Chu
ICLR 2026 — paper
2025
Latent Adversarial Reflection for LLM Jailbreaking thumbnail

Latent Adversarial Reflection for LLM Jailbreaking

Ran Li, Hao Wang, Chengzhi Mao
NeurIPS 2025 — paper
Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision thumbnail

Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision

Chenshuang Zhang, Kang Zhang, Joon Son Chung, In So Kweon, Junmo Kim, Chengzhi Mao
NeurIPS 2025 — paper
EDITLORD: Learning Code Transformation Rules for Code Editing thumbnail

EDITLORD: Learning Code Transformation Rules for Code Editing

Weichen Li, Albert Jan, Baishakhi Ray, Junfeng Yang, Chengzhi Mao, Kexin Pei
ICML 2025 — paper
Diversity Helps Jailbreak Large Language Models thumbnail

Diversity Helps Jailbreak Large Language Models

Weiliang Zhao, Daniel Ben-Levi, Wei Hao, Junfeng Yang, Chengzhi Mao
NAACL 2025 (Oral) — paper
2024
SelfIE: Self-Interpretation of Large Language Model Embeddings thumbnail

SelfIE: Self-Interpretation of Large Language Model Embeddings

Haozhe Chen, Carl Vondrick, Chengzhi Mao
ICML 2024 — project
Raidar: geneRative AI Detection viA Rewriting thumbnail

Raidar: geneRative AI Detection viA Rewriting

Chengzhi Mao, Carl Vondrick, Hao Wang, Junfeng Yang
ICLR 2024 — paper
2023
Doubly Right Object Recognition: A Why Prompt for Visual Rationales thumbnail

Doubly Right Object Recognition: A Why Prompt for Visual Rationales

Chengzhi Mao, Revant Teotia, Amrutha Sundar, Sachit Menon, Junfeng Yang, Xin Wang, Carl Vondrick
CVPR 2023 — arXiv / dataset / code
Shadows Shed Light on 3D Objects thumbnail

Shadows Shed Light on 3D Objects

Ruoshi Liu, Sachit Menon, Chengzhi Mao, Dennis Park, Simon Stent, Carl Vondrick
CVPR 2023 — arXiv / dataset / code
2022
Real-Time Neural Voice Camouflage thumbnail

Real-Time Neural Voice Camouflage

Mia Chiquier, Chengzhi Mao, Carl Vondrick
ICLR 2022 (Oral) — arXiv / code / Science / talk
Discrete Representations Strengthen Vision Transformer Robustness thumbnail

Discrete Representations Strengthen Vision Transformer Robustness

Chengzhi Mao, Lu Jiang, Mostafa Dehghani, Carl Vondrick, Rahul Sukthankar, Irfan Essa
ICLR 2022 — arXiv / code / cite / talk
2021
Adversarial Attacks are Reversible with Natural Supervision thumbnail

Adversarial Attacks are Reversible with Natural Supervision

Chengzhi Mao, Mia Chiquier, Hao Wang, Junfeng Yang, Carl Vondrick
ICCV 2021 — arXiv / code / cite / talk
Generative Interventions for Causal Learning thumbnail

Generative Interventions for Causal Learning

Chengzhi Mao, Amogh Gupta*, Vikram Nitin*, Baishakhi Ray, Shuran Song, Junfeng Yang, Carl Vondrick
CVPR 2021 — arXiv / code / cite / talk

Teaching

Contact

Department of Computer Science, Rutgers University
Email: chengzhi.mao@rutgers.edu

Visitors