Niao He
- AmorLIP: Efficient Language-Image Pretraining via Amortization
- Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning
- Natural Gradient VI: Guarantees for Non-Conjugate Models
- PoLAR: Polar-Decomposed Low-Rank Adapter Representation
- Scalable Neural Incentive Design with Parameterized Mean-Field Approximation
- Zeroth-Order Optimization Finds Flat Minima