🔎 Research Interests

My work centers on discovering intrinsic structural priors in complex systems and using them to guide deep representation and generation. Current directions:

a) AI for Science

• Cross-omics modeling along the Central Dogma (DNA ↔ RNA ↔ protein)
• Single-cell dynamics and context-aware biological representation
• Structure-based drug design and molecular modality expansion

b) Hierarchical Part-Whole Representation

• Capsule networks for unsupervised part discovery
• 3D face modeling beyond 3DMM
• 2D→2.5D→3D pathway in neural networks (Marr-inspired)

c) Unified Dynamics on Data Manifolds

• ODE discovery, RNA velocity inference, and AI-generated video detection
• Manifold decoupling for cross-task dynamics modeling