高斯流匹配调度:对采样与训练的影响
Gaussian Flow-Matching Schedules: Implications for Sampling and Training
浏览论文内容
中文总结 AI 辅助
本文研究流匹配调度对采样与训练的影响,将调度分解为方差路径和分解,推导采样精度界及最优训练分解。
中文摘要 AI 辅助
流匹配调度同时影响采样动力学和回归目标的方差。对于中心化可交换高斯分布,我们证明方向依赖的调度可分解为两个独立的设计选择:方差路径(完全决定中间分布和概率流)和分解(保持该流不变,同时控制不可约回归方差)。在采样方面,我们分析了有限步欧拉精度,并推导了精确N步采样的必要漂移界,连接了测地线路径和对数路径。在训练方面,对于任意固定路径,我们推导了闭式分解,这些分解要么最小化时间平均回归方差,要么使其沿路径保持恒定。
英文摘要
Flow-matching schedules affect both sampling dynamics and the variance of the regression target. For centered commuting Gaussians, we show that a direction-dependent schedule decomposes into two independent design choices: a variance path, which fully determines the intermediate laws and probability flow, and a factorization, which leaves this flow unchanged while controlling irreducible regression variance. On the sampling side, we analyze finite-step Euler accuracy and derive a necessary drift bound for exact N -step sampling, connecting the geodesic and the logarithmic path. On the training side, for any fixed path, we derive closed-form factorizations that either minimize time-averaged regression variance or make it constant along the path.
发表机构
- CNRS(法国国家科学研究中心)
- ENS Paris(巴黎高等师范学院)
- Capital Fund Management(资本基金管理公司)
- Laboratoire de Mathématiques d’Orsay(奥赛数学实验室)
- Université Paris-Saclay(巴黎萨克雷大学)
- Institut Universitaire de France(法国大学研究院)
- Courant Institute of Mathematical Sciences, New York University(纽约大学库朗数学科学研究所)
机构由 AI 辅助整理,请以论文原文为准。