Young Researcher · Shanghai AI Laboratory
Mingxiao Li
I work on generative and understanding models for AI for Science, with a continuing interest in visual generation and multimodal intelligence.
My current work at Shanghai AI Laboratory focuses on AI for Science discovery, with an emphasis on molecular generation, unified molecular tokenization, and reasoning models for scientific problems. In parallel, I continue to develop visual generation models for controllable video synthesis and multimodal understanding.
Before joining Shanghai AI Laboratory, I was a postdoctoral researcher at KU Leuven, where my research was guided by Prof. Marie-Francine Moens.
Research
I study how generative models can discover, represent, and reason about complex scientific and visual worlds.
- AI for Science discoveryMolecular generation, unified molecular tokenization, and foundation models for scientific exploration.
- Scientific reasoning modelsModels that connect evidence, language, and structured representations to support reliable scientific reasoning.
- Visual generationDiffusion and multimodal models for controllable images, video, motion, and dynamic visual experiences.
Education
Young Researcher
Shanghai AI Laboratory · AI for Science Center
Postdoctoral Researcher
Computer Science · KU Leuven
PhD in Computer Science
KU Leuven · summa cum laude
Master of Artificial Intelligence
KU Leuven
Master of Theoretical Chemistry & Computational Modeling
KU Leuven
Bachelor of Material Physics
East China University of Science and Technology
Selected publications
Full list on Google Scholar ↗- NeurIPS 2025
Consistent Story Generation: Unlocking the Potential of Zigzag Sampling. [Paper]
- ICML 2025
DCTdiff: Intriguing Properties of Image Generative Modeling in the DCT Space. [Paper]
- AAAI 2025 · Oral
NeuroCine: Decoding Vivid Video Sequences from Human Brain Activities. [Paper]
- ECCV 2024
Animate Your Motion: Turning Still Images into Dynamic Videos. [Paper]
- ICLR 2024
Alleviating Exposure Bias in Diffusion Models through Shifted Time Steps. [Paper]
- NeurIPS 2023
Contrast, Attend and Diffuse to Decode High-Resolution Images from Brain Activities. [Paper]
- ICLR 2024
Elucidating the Exposure Bias in Diffusion Models. [Paper]
- AAAI 2023 · Oral
Layout-Aware Dreamer for Embodied Visual Grounding. [Paper]
Technical notes
Open the notes archive ↗I write about the techniques behind my work: diffusion, sampling, molecular representations, multimodal reasoning, and the practical craft of research engineering.
Service & awards
Academic service
Area Chair · ACL Rolling Review (2025.02)
Workshop organizer · AAAI 2024 AIBED
Reviewer for ICML, CVPR, ICLR, NeurIPS, AAAI, EMNLP, ACL, ECAI, ECML, and EACL.
Awards
Summa cum laude PhD
Erasmus Mundus Full Scholarship
Chinese Government Scholarship
Contact
I am always happy to hear about thoughtful questions, unusual ideas, and potential collaborations.
eric.lee.xiao@gmail.com