Tag
This paper introduces Newton Matching, a unified framework for fine-tuning and sampling in generative models, which addresses limitations of existing methods by treating learning as an iterative optimization process and leveraging conditional-matching structure.