Brain-DiT: A Universal Multi-state fMRI Foundation Model with Metadata-Conditioned Pretraining
MICCAI 2026 (🥇 Best Paper Award & Young Scientist Award Shortlist). GitHubPoster
About
I am Junfeng Xia (夏俊锋), a master's student in Biomedical Engineering at Southern University of Science and Technology, advised by Prof. Quanying Liu (刘泉影). I expect to graduate in June 2027 and am applying for Fall 2027 Ph.D. programs. Before joining SUSTech, I received my B.Eng. in Computer Science and Technology from Zhengzhou University, where I was advised by Prof. Qidong Liu (刘起东).
I am especially interested in three questions:
- How can robust neural representations be learned from heterogeneous brain signals?
- How should multimodal neural representations be fused?
- How can multimodal neural representations become useful brain knowledge?
News
- Sep 2026🎉 BrainWorld was accepted to NeurIPS 2026 as an oral presentation.
- Sep 2026🎉 FlatClip was accepted to NeurIPS 2026 as a poster presentation.
- Sep 2026
was released on arXiv. We strongly recommend this work; it presents a novel and intriguing reconstruction-based information decomposition method for constructing functional brain atlases. - Sep 2026🎉 Our paper Brain-DiT was shortlisted at MICCAI 2026 for the 🥇 Best Paper Award and Young Scientist Award TOP 0.5%.
- Jun 2026Brain-DiT was accepted by MICCAI 2026. Looking forward to visiting Strasbourg.
- Jun 2026BrainWorld was released on arXiv.
- Jun 2026FlexiBrain was accepted by ECCV 2026.
- Apr 2026Omni-fMRI was accepted by ICML 2026.
- Jun 2025SLIM-Brain was released on arXiv.
- Jun 2025I was selected for the Croucher Summer Course in Computational Neuroscience at the Chinese University of Hong Kong.
- May 2025Our work on ultralow-power ion-gel nanofiber artificial synapses for enhanced working memory was published in Advanced Materials.
- Sep 2024Our work on A Genetic Algorithms for Optimizing Structural Brain Network Across Cognitive Tasks was published in China Automation Congress.
- Jun 2024Our work on Uncovering Cognitive Taskonomy through Transfer Learning in Masked Autoencoder-based fMRI Reconstruction was published in HBAI,IJCAI(Oral).
- Jun 2023I began research training at Southern University of Science and Technology as a visiting student.
Research Line
Selected Publications
* denotes equal contribution.
: Reconstruction-Based Information Decomposition for Functional Brain Atlas Construction
arXiv, 2026.
BrainWorld: A Structural-Prior-Conditioned Generative Model for Whole-Brain 4D fMRI Dynamics
NeurIPS, 2026 (Oral). GitHub
FlatClip: Reusing Image Foundation Models for fMRI Representation Learning via Cortical Flatmaps
NeurIPS, 2026 (Poster).
BrainTaskonomy: Learning How to Pretrain and What to Transfer in fMRI Foundation Models
arXiv:2609.10518, 2026.
SLIM-Brain: A Data- and Training-Efficient Foundation Model for fMRI Data Analysis
arXiv:2512.21881, 2025. GitHub
Advanced Materials, 37(16), 2419013, 2025.
A Genetic Algorithms for Optimizing Structural Brain Network Across Cognitive Tasks
China Automation Congress, 5210-5215, 2024.
Manuscripts Under Review and In Preparation
* denotes equal contribution.
Uncovering information hubs in human brain dynamics through prediction-based decomposition
Manuscript in preparation.
A Scaling Study for fMRI Foundation Models
Under review.
CoFlu-Brain: An fMRI Connectivity Foundation Model with Cofluctuation-Based Pretraining
Under review.
fMRI Guidance for EEG Diffusion Representation Learning and Synthetic Data Augmentation
Under review.
Books
- Human Brain Intelligence and Artificial Intelligence. Tsinghua University Press, 2025.
Service
Reviewing
- Conference Reviewer for ICLR.
Teaching
- Teaching Assistant, Machine Learning and Medical Engineering Applications, SUSTech, Shenzhen, 2025.
Academic Activities
- Visiting Student, Neural Computing and Control Lab, Southern University of Science and Technology, 2023-2024.
- Selected participant, Croucher Summer Course in Computational Neuroscience, Chinese University of Hong Kong, Jun. 2025.