Wenhao Ye | NeuroAI & fMRI Foundation Models

BRAIN SCIENCE × MACHINE LEARNING

About me

I study how we can learn useful representations from the human brain.

I am an undergraduate in Biomedical Engineering at Shenzhen University, and a visiting research student at the Neural Computing and Control (NCC) Lab, SUSTech, advised by Prof. Quanying Liu.

My research focuses on NeuroAI, particularly foundation models and representation learning for fMRI. I am also exploring trustworthy AI systems with JEPA and uncertainty-aware learning.

fMRI foundation modelsRepresentation learningBrain–computer interfaces

Publications & manuscripts

8 works

More work from our team: fMRI Atlas

* Equal contribution · † Corresponding author

Omni-fMRI

ICML 2026Co-first author

Omni-fMRI: A Universal Atlas-Free fMRI Foundation Model

M. Wang*, Wenhao Ye*, J. Xia, J. Zhang, X. Pan, M. Xu, H. Deng, H. Wen^†, Q. Liu^†

Dynamic patching enables efficient, atlas-free representation learning directly from voxel-level fMRI.

Paper Code

FlexiBrain

ECCV 2026Co-first author

FlexiBrain: Resolution-Agnostic Voxel-Level Encoding for Native fMRI

M. Wang*, Wenhao Ye*, J. Xia*, M. Xu, H. Wen^†, Q. Liu^†

Physical-unit patching and Mamba-JEPA learn native-space fMRI representations across different acquisition resolutions.

Paper Code

Overview of the scaling study for fMRI foundation models

Scaling fMRI models

Preprint · 2026First author

A Scaling Study for fMRI Foundation Models

Wenhao Ye, X. Pan, J. Xia, J. Zhang, J. Yu, M. Wang^†, Q. Liu^†

A controlled scaling study reveals how data size, model capacity, and training duration jointly affect fMRI performance.

Paper Code

BrainWorld

NeurIPS 2026Second authorOral · Top 0.36%

BrainWorld: A Structural-Prior World Model for Whole-Brain 4D fMRI Generation

J. Xia, Wenhao Ye, J. Zhang, X. Pan, M. Wang^†, Q. Liu^†

Structural MRI guides the generation of whole-brain fMRI sequences for modeling brain dynamics and learning transferable representations.

Paper Code

FlatCLIP

NeurIPS 2026Second author

FlatCLIP: A Geometry-Aware Surface-Level Baseline for fMRI Representation Learning

M. Wang, Wenhao Ye, Z. Ning, J. Zuo, J. Xia, H. Wen^†, Q. Liu^†

A frozen image foundation model extracts fMRI representations from geometry-aware cortical flatmaps.

Paper Code

Brain-DiT

MICCAI 2026Second authorBest Paper & Young Scientist Award shortlist · Top 0.5%

Brain-DiT: A Universal Multi-state fMRI Foundation Model with Metadata-Conditioned Pretraining

J. Xia, Wenhao Ye, X. Pan, X. Shen, M. Wang^†, Q. Liu^†

Metadata-conditioned diffusion pretraining learns multi-scale fMRI representations across diverse brain states.

Paper Code

BrainTaskonomy

Preprint · 2026Second author

BrainTaskonomy: Learning How to Pretrain and What to Transfer in fMRI Foundation Models

J. Xia, Wenhao Ye, J. Zhang, J. Zuo, M. Wang^†, Q. Liu^†

Measured relationships between fMRI domains and tasks guide pretraining curricula and budget-aware transfer learning.

Paper

SLIM-Brain

Preprint · 2025Second author

SLIM-Brain: A Data- and Training-Efficient Foundation Model for fMRI Data Analysis

M. Wang*, J. Xia*, Wenhao Ye, E. Liu, K. Peng, J. Feng, Q. Liu^†, H. Wen^†

Salient-window selection and hierarchical encoding reduce the data and memory costs of voxel-level fMRI modeling.

Paper Code

Education & experience

2025 — Present

Southern University of Science and Technology

Visiting Research Student · NCC Lab

Advisor: Prof. Quanying Liu · Since July 2025

2023 — 2027

Shenzhen University

B.Eng. in Biomedical Engineering

Expected June 2027