发表机构
Flatiron Institute; Technical University of Denmark; Courant Institute, New York University（弗拉蒂隆研究所; 丹麦技术大学; 纽约大学柯朗研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出SGFlow方法，绕过流图学习的可逆性与迭代微分难题，在CIFAR基准上10采样步时FID最优，其他步数具竞争力且有驻点保证。
AI 中文摘要
扩散模型和基于流的模型得益于简单的回归损失，但推理因采样需要积分而产生显著开销。一致性模型通过直接学习常微分方程（ODE）轨迹上的流图解决该问题，在单步与多步方法间开辟了设计空间。然而现有方法存在计算挑战，如需要模型逆或通过迭代模型调用的反向传播，且未总能证明所需ODE流图是损失的解。本文提出SGFlow，一种学习流图的方法，绕过显式可逆性约束和通过模型迭代的昂贵微分。SGFlow训练模型从头计算ODE解和隐含速度，遵循非保守动力学，其驻点为所需流图。在CIFAR图像基准上，无单一方法在每一步数都取得最佳FID：SGFlow在10个采样步时取得最佳FID，在其他步数时与流匹配（flow matching）、Meanflow、拉格朗日图匹配（Lagrangian map matching）有竞争力，且是唯一具有基于停止梯度（stopgrad）动力学的驻点保证的方法。
英文摘要
Diffusion and flow-based models benefit from simple regression losses, but inference incurs significant overhead because sampling requires integration. Consistency models address this by directly learning the flow maps along the ODE trajectory, opening a design space between one-step and many-step approaches. However, existing methods face computational challenges such as requiring model inverses or backpropagation through iterated model calls, and do not always prove that the desired ODE flow map is a solution to the loss. We introduce SGFlow, an approach for learning flow maps that bypasses explicit invertibility constraints and expensive differentiation through model iteration. SGFlow trains a model to compute both the ODE solutions and the implied velocity from scratch by following non-conservative dynamics with a stationary point at the desired flow map. On the CIFAR image benchmark, no single method attains the best FID at every step count: SGFlow attains the best FID at 10 sampling steps and remains competitive with flow matching, Meanflow, and Lagrangian map matching at other step counts, while being the only one with a proven stationary-point guarantee for its stopgrad-based dynamics.
CommentsICLR 2026
