MS student, Korea Advanced Institute of Science & Technology
3 papers at NeurIPS 2025
Inference-time scaling for flow models
We propose a novel regularization loss that enforces standard Gaussianity, encouraging samples to align with a standard Gaussian distribution.