About me

I am Dinghuai Zhang (张鼎怀), a senior researcher at Microsoft Research New England. I have received my PhD training at Mila in Prof. Bengio’s group as well as Aaron’s Army (aka A.A.). During my PhD, I’ve interned at Apple MLR and FAIR labs. Previously, I was an undergraduate in School of Mathematical Sciences at Peking University, working with Prof. Zhanxing Zhu and Prof. Bin Dong.

I am always open to possible discussion and research collaboration with junior students. Students from underrepresented groups are strongly encouraged to reach out! If you are interested, feel free to contact me by email, facebook, or wechat.

Research interests

I aim to tackle multidisciplinary problems with multimodal generative models.

Probabilistic methods: generative models, amortized inference, uncertainty modeling
Exploration with structure: RL, search, black-box optimization, active learning
Scientific design & discovery as probabilistic inference

I’ve also conducted AI safety research (distribution shift, causal representation learning).

Publications

Efficient Diversity-Preserving Diffusion Alignment via Gradient-Informed GFlowNets [project page]
Zhen Liu, Tim Zhenzhong Xiao*, Weiyang Liu*, Yoshua Bengio, Dinghuai Zhang
ICLR 2025

Improving GFlowNets for Text-to-Image Diffusion Alignment [code]
Dinghuai Zhang, Yizhe Zhang, Jiatao Gu, Ruixiang Zhang, Josh Susskind, Navdeep Jaitly, Shuangfei Zhai
TMLR 2024

Latent Energy-Based Odyssey: Black-Box Optimization via Expanded Exploration in the Energy-Based Latent Space
Peiyu Yu*, Dinghuai Zhang*, Hengzhi He*, Xiaojian Ma, Ruiyao Miao, Yifan Lu, Yasi Zhang, Deqian Kong, Ruiqi Gao, Jianwen Xie, Guang Cheng, Ying Nian Wu
In submission

Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization
Dinghuai Zhang, Ricky Tian Qi Chen, Cheng-Hao Liu, Aaron Courville, Yoshua Bengio.
12th International Conference on Learning Representations (ICLR 2024)

Let the Flows Tell: Solving Graph Combinatorial Optimization Problems with GFlowNets
Dinghuai Zhang, Hanjun Dai, Nikolay Malkin, Aaron Courville, Yoshua Bengio, Ling Pan.
37th Conference on Neural Information Processing Systems (NeurIPS 2023 spotlight)

Stochastic Generative Flow Networks [openreview]
Ling Pan*, Dinghuai Zhang*, Moksh Jain, Longbo Huang, Yoshua Bengio.
39th Conference on Uncertainty in Artificial Intelligence (UAI 2023 spotlight)

Distributional GFlowNets with Quantile Flows [slides]
Dinghuai Zhang*, Ling Pan*, Ricky Tian Qi Chen, Aaron Courville, Yoshua Bengio.
Transactions on Machine Learning Research (TMLR)

Unifying Generative Models with GFlowNets and Beyond
Dinghuai Zhang, Ricky Tian Qi Chen, Nikolay Malkin, Yoshua Bengio.
ICML 2022 Beyond Bayes workshop.

Latent State Marginalization as a Low-cost Approach for Improving Exploration [openreview] [slides]
Dinghuai Zhang, Aaron Courville, Yoshua Bengio, Qinqing Zheng, Amy Zhang, Ricky Tian Qi Chen.
11th International Conference on Learning Representations (ICLR 2023)

Predictive Inference with Feature Conformal Prediction [openreview] [poster]
Jiaye Teng*, Chuan Wen*, Dinghuai Zhang*, Yoshua Bengio, Yang Gao, Yang Yuan.
11th International Conference on Learning Representations (ICLR 2023)

Generative Flow Networks for Discrete Probabilistic Modeling [poster] [slides]
Dinghuai Zhang, Nikolay Malkin, Zhen Liu, Alexandra Volokhova, Aaron Courville, Yoshua Bengio.
39th International Conference on Machine Learning (ICML 2022)

Building Robust Ensembles via Margin Boosting [poster] [slides]
Dinghuai Zhang, Hongyang Zhang, Aaron Courville, Yoshua Bengio, Pradeep Ravikumar, Arun Sai Suggala.
39th International Conference on Machine Learning (ICML 2022)

Unifying Likelihood-free Inference with Black-box Optimization and Beyond [openreview] [slides]
Dinghuai Zhang, Jie Fu, Yoshua Bengio, Aaron Courville.
10th International Conference on Learning Representations (ICLR 2022 spotlight)

Can Subnetwork Structure be the Key to Out-of-Distribution Generalization? [poster] [slides]
Dinghuai Zhang, Kartik Ahuja, Yilun Xu, Yisen Wang, Aaron Courville.
38th International Conference on Machine Learning (ICML 2021 long talk)

Neural Approximate Sufficient Statistics for Implicit Models [openreview] [poster] [slides] [Michael’s slides]
Yanzhi Chen*, Dinghuai Zhang*, Michael Gutmann, Aaron Courville, Zhanxing Zhu.
9th International Conference on Learning Representations (ICLR 2021 spotlight)

Black-Box Certification with Randomized Smoothing: A Functional Optimization Based Framework [poster] [slides]
Dinghuai Zhang*, Mao Ye*, Chengyue Gong* , Zhanxing Zhu, Qiang Liu.
34th Conference on Neural Information Processing Systems (NeurIPS 2020)

Informative Dropout for Robust Representation Learning: A Shape-bias Perspective [slides] [zhihu]
Baifeng Shi*, Dinghuai Zhang*, Qi Dai, Zhanxing Zhu, Yadong Mu, Jingdong Wang.
37th International Conference on Machine Learning (ICML 2020)

You Only Propagate Once: Accelerating Adversarial Training via Maximal Principle [poster] [slides] [zhihu]
Dinghuai Zhang*, Tianyuan Zhang*,Yiping Lu*, Zhanxing Zhu, Bin Dong.
33rd Conference on Neural Information Processing Systems (NeurIPS 2019)

( * denotes equal contribution)

Activities

Organizer

Safe Generative AI Workshop at NeurIPS 2024.
Structured Probabilistic Inference & Generative Modeling (SPIGM) workshop at ICML 2023 and ICML 2024.

Reviewer

JMLR, TPAMI, ICML, NeurIPS, ICLR, AISTATS, UAI, CLeaR, EMNLP.

Miscs

I used to be a huge fan of Daido Moriyama and Henri Cartier-Bresson (street photography pioneers).
I paint Chinese calligraphy well since a child.
My Erdős number is 3.