BriefGPT.xyz
Mar, 2025
归纳时刻匹配
Inductive Moment Matching
HTML
PDF
Linqi Zhou, Stefano Ermon, Jiaming Song
TL;DR
本研究解决了扩散模型和流匹配在推理过程中的速度慢以及在少步模型中常导致的不稳定性和大量调优问题。提出的归纳时刻匹配(IMM)是一种新型生成模型,通过单阶段训练程序实现一到少步取样,且在多个超参数和标准模型架构下保证分布级收敛,取得了在ImageNet-256x256和CIFAR-10数据集上的最新状态。
Abstract
Diffusion Models
and Flow Matching generate high-quality samples but are slow at inference, and distilling them into few-step models often leads to instability and extensive tuning. To resolve these trade-offs, we propose
→