自适应重参数化时间 ART：用强化学习优化扩散模型采样时间步分配
Adaptive Reparametrized Time for Score-Based Diffusion Sampling
AI 导读
一项被 JMLR 接收的研究提出自适应重参数化时间（ART），把采样时钟速度作为控制量，将扩散采样的时间步分配学习转化为连续时间控制问题。配套的 ART-RL 用高斯策略的随机化形式求解，证明二者在优化器层面等价，并给出可实现的 actor-critic 更新方法。
正文 · 原文
Abstract:We study timestep allocation for score-based diffusion sampling, where a learned reverse-time dynamics is discretized on a finite grid. Uniform and hand-crafted schedules are standard choices, but they rely on fixed prescriptions and can therefore be suboptimal. To address this limitation, we propose Adaptive Reparameterized Time (ART), a continuous-time control formulation that learns a time change by treating the speed of the sampling clock as the control, so that a uniform grid on the learned clock induces adaptive timesteps in the original diffusion time. Based on a leading-order Euler error surrogate, ART provides a principled objective for allocating timesteps along the sampling trajectory. To solve this deterministic control problem, we introduce ART-RL, an auxiliary randomized formulation with Gaussian policies that turns schedule learning into a continuous-time reinforcement learning problem. We prove that the randomized ART-RL formulation is equivalent to ART at the optimizer level, in the sense that its optimal Gaussian policy recovers the optimal ART time-warping rate through its mean. We further establish policy evaluation and policy improvement characterizations and derive trajectory-based moment identities that yield implementable actor--critic updates for learning the schedule. Across experiments ranging from controlled low-dimensional settings to image generation, ART-RL can be plugged into existing diffusion samplers by changing only the timestep grid, consistently improving sample quality over strong baseline schedules at matched budgets while leaving the rest of the sampling pipeline unchanged. The learned schedules also exhibit broad generalization, transferring without retraining across sampling budgets, datasets, solvers, pipelines, and representation spaces.
| Comments: | 37 pages, 14 figures, 8 tables. Accepted for publication in the Journal of Machine Learning Research (JMLR) |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Systems and Control (eess.SY); Optimization and Control (math.OC) |
| Cite as: | arXiv:2607.02137 [cs.LG] |
| (or arXiv:2607.02137v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2607.02137 arXiv-issued DOI via DataCite |
Submission history
From: Yilie Huang [view email]
[v1]
Thu, 2 Jul 2026 13:13:19 UTC (24,062 KB)
[v2]
Fri, 3 Jul 2026 02:14:28 UTC (24,062 KB)
[v3]
Wed, 30 Sep 2026 09:17:14 UTC (24,062 KB)
[v4]
Tue, 10 Nov 2026 14:15:47 UTC (24,062 KB)
来源：arXiv cs.AI · arxiv.org