About me

I am a PhD student at Université de Montréal and Mila (Quebec AI Institute), supervised by Aaron Courville. Previously, I was a visiting researcher in the AIM program at FAIR, Meta, where I worked with Adriana Romero-Soriano and Michal Drozdzal.

My research focuses on image and video generation and representation learning. I am currently exploring how world models can learn across modalities to better understand the physical world and support reasoning and planning in real-world environments.

Keywords: Image and video generation, representation learning, world models, multimodal learning, reasoning and planning.

Services: Reviewer for ICLR 2024, ICML 2025, NeurIPS 2025, EMNLP 2025, CVPR 2026, ICML 2026, and ICLR 2027.
Organizer of the World Modeling Workshop and the Mila World Models Reading Group.

News

  • Oct. 2026 FeaTok: our paper on discrete-token image generation is out! FeaTok learns discrete image tokens from pretrained visual features and uses a separate decoder to reconstruct images, improving the trade-off between generation quality and computational cost. Stay tuned for model checkpoints and code!
  • Oct. 2026 Attended the EDPM workshop in New York City! Thank you to the organizers of Emerging Directions in Probabilistic Modeling for hosting a wonderful workshop filled with excellent talks and engaging discussions.
  • May 2026 Awarded the FRQNT PhD Scholarship! I am honored to receive support from the Fonds de recherche du Québec – Nature et technologies (FRQNT) for my PhD research on generative models, world modeling, and how AI systems represent the diversity of the real world.
  • Feb. 2026 Our paper was accepted to CVPR 2026! The study shows how signals from world models such as V-JEPA 2 can improve physical plausibility in generated videos at inference time, across sampling methods.
  • Feb. 2026 Our paper was accepted to ICLR 2026! The study examines how prompt complexity affects the diversity, quality, and consistency of generated images, and how these images compare with real data.
  • Feb. 2026 First World Modeling Workshop at Mila! I was delighted to help organize the inaugural workshop in Montréal. Explore the talk recordings and poster list on our website.
  • Dec. 2025 Attended NeurIPS 2025 in San Diego! Explore Chamfer Guidance, our work on using Chamfer distance to align the output distribution of pretrained text-to-image models with a reference dataset.

View all news →

Contact

Mila (Quebec AI Institute),

6666 Rue Saint-Urbain,

Montréal, QC H2S 3H1, Canada

xiaofng [DOT] zhang [AT] gmail [DOT] com



Let’s connect

I’m always happy to discuss research and exchange ideas. Check the calendar below for my availability, or book a meeting directly.


Daniel lewis
quote icon

Daniel lewis

Resume

Education

Experience

  1. Natural Language Processing Engineer    Pattern Recognition Center, Wechat AI, Tencent

    Dec. 2021 - Jun. 2022    supervised by Dr. Yikang Shen

    Worked on two research topics. Firstly, we propose a new attention mechanism called Mixture of Attention Heads (MoA), which is a combination of Mixture of Experts and Multi-head Attention Mechanism. MoA permits easily scale up the model capacity while constraining the same level of computational cost. We achieved better performance on Machine Translation and Language Modeling tasks compared to SOTA baselines. Secondly, we work on the model editing task in order to efficiently correct the false prediction of a pre-trained language model. We propose a new task named sequential editing task and conceive a novel way to edit the stored knowledge in the pre-trained language model. We add a few new neurons in the last Feed Forward layer of Transformer architecture. The experimental result show that our method outperforms other strong baselines.

Publication

Blog

News

  • Oct. 2026 FeaTok: our paper on discrete-token image generation is out! FeaTok learns discrete image tokens from pretrained visual features and uses a separate decoder to reconstruct images, improving the trade-off between generation quality and computational cost. Stay tuned for model checkpoints and code!
  • Oct. 2026 Attended the EDPM workshop in New York City! Thank you to the organizers of Emerging Directions in Probabilistic Modeling for hosting a wonderful workshop filled with excellent talks and engaging discussions.
  • May 2026 Awarded the FRQNT PhD Scholarship! I am honored to receive support from the Fonds de recherche du Québec – Nature et technologies (FRQNT) for my PhD research on generative models, world modeling, and how AI systems represent the diversity of the real world.
  • Feb. 2026 Our paper was accepted to CVPR 2026! The study shows how signals from world models such as V-JEPA 2 can improve physical plausibility in generated videos at inference time, across sampling methods.
  • Feb. 2026 Our paper was accepted to ICLR 2026! The study examines how prompt complexity affects the diversity, quality, and consistency of generated images, and how these images compare with real data.
  • Feb. 2026 First World Modeling Workshop at Mila! I was delighted to help organize the inaugural workshop in Montréal. Explore the talk recordings and poster list on our website.
  • Dec. 2025 Attended NeurIPS 2025 in San Diego! Explore Chamfer Guidance, our work on using Chamfer distance to align the output distribution of pretrained text-to-image models with a reference dataset.
  • Sep. 2025 Attended the PAISS summer school in Grenoble! A wonderful three days in Grenoble. Browse the highlights on the PAISS website.
  • Aug. 2025 Our paper was accepted to Findings of EMNLP 2025! CulturalFrames evaluates how leading open- and closed-source text-to-image models respond to culturally informed prompts and capture implicit and explicit cultural information.
  • Mar. 2025 Awarded the EDI in Research Scholarship! I am grateful to Mila for supporting my research through this scholarship.
  • Dec. 2024 Attended NeurIPS 2024 in Vancouver! I presented my work on bias analysis in unconditional image generation models.
  • Sep. 2024 Joined FAIR, Meta, as a visiting researcher!
  • Nov. 2023 Our paper was accepted to Findings of EMNLP 2023! Read our paper on pretrained model selection.
  • Jan. 2023 Our paper was accepted to ICLR 2023! Read our paper on knowledge updates and model editing for large language models.
  • Sep. 2022 Two papers were accepted to EMNLP 2022! Explore our papers on mixture-of-experts attention architectures and on prompt learning and meta-learning.