MGPO: Manifold-Guided Diffusion Alignment for Task-Aware Dataset Distillation
Published in arXiv, 2026
This paper introduces MGPO, a manifold-guided approach to diffusion alignment for task-aware dataset distillation. The method leverages manifold geometry to guide the diffusion process, improving the quality of distilled datasets for specific downstream tasks.
Recommended citation: Yunyi Chen, Chenru Wang, Xinyi Ye, Zexin Zheng, Chi Zhang. (2026). "MGPO: Manifold-Guided Diffusion Alignment for Task-Aware Dataset Distillation."