BriefGPT.xyz
Jan, 2023
基于ODE的生成模型轨迹曲率最小化
Minimizing Trajectory Curvature of ODE-based Generative Models
HTML
PDF
Sangyun Lee, Beomsu Kim, Jong Chul Ye
TL;DR
该论文提出了一种基于 curvature minimization 的有效的 ODE / SDE 学习方法,能够在不需要多次神经网络评估的情况下减少生成轨迹的曲率,从而减少采样成本并保持了竞争性性能。
Abstract
Recent
ode/sde-based generative models
, such as diffusion models and flow matching, define a generative process as a time reversal of a fixed forward process. Even though these models show impressive performance on large-scale datasets,
→