Researcher profile
Research Staff · 科研人员
South China University of Technology
Learning robust representations for molecular discovery, computational pathology, and digital humans.
About
Yujian is a research staff member in the Xu Group at South China University of Technology, where he completed his MSc in Biomedical Engineering under the supervision of Prof. Shidang Xu in 2026. His research spans molecular discovery, computational pathology, and digital humans, with a shared focus on learning useful representations under structural uncertainty, limited computation, and geometric constraints.
In molecular discovery, he develops physics-informed representations for protein-ligand interaction prediction and molecular design. In computational pathology, he studies how local tissue detail and slide-level context can inform each other across magnifications. His digital-human research explores 3D Gaussian avatars and speech-driven animation, including reconstruction from uncalibrated views and coordinated facial and upper-body motion.
His broader interests include embodied intelligence for autonomous scientific experimentation. He received the National Scholarship for Graduate Students during his master’s studies.
Research questions
Learning from electron-density point clouds and molecular graphs to support protein-ligand interaction prediction under structural uncertainty.
Using spatial correspondence across magnifications and cross-scale reconstruction to learn from gigapixel pathology images.
Combining 3D reconstruction and speech-driven animation while accounting for uncertain camera poses and facial geometry.
Current work
Research direction
Physics-informed molecular learning for drug discovery, including protein-ligand interaction prediction and generative molecular design. E-CloudBind is a published example of this direction.
My role: Molecular representation and model development; co-first author of E-CloudBind.
Research direction
Efficient learning from tissue images at multiple magnifications, connecting local evidence with slide-level representations through cross-scale fusion and super-resolution supervision.
My role: First author of MPFusion-MIL and co-first author of PyraE2E.
Research direction
Animatable 3D Gaussian head reconstruction and audio-driven motion, with an emphasis on geometric consistency and practical deployment.
My role: Model and algorithm development for AnyAvatar, MoGaFace, and SyncAnimation.
Contribution profile
Selected work
2026
Nat. Commun., 17(1), 7424 (2026)
2026
ACM MM, 2026 — Accepted
2026
ECCV, 2026 — Accepted
2026
ACM MM, 2026 — Accepted
2025
IJCAI-25, 1657–1665 (2025)