arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.37146cs.LG
收藏

用于扩散引导的变分最优控制

Variational Optimal Control for Diffusion Guidance

Westlake University(西湖大学);Shandong University(山东大学);Shanghai Jiao Tong University(上海交通大学);University of Science and Technology of China(中国科学技术大学)

Ruoyu Wang, Xinyi Ye, Yunshuo Shi, Liangyu Yuan, Beier Zhu, Chi Zhang

AI总结:

该研究提出变分控制框架VOG与DVOC,通过几何原理关联采样引导与模型自适应,在ImageNet等任务中显著降低FID并提升评估分数。

摘要:

我们提出一种变分控制框架,通过共享雅可比拉回映射将采样时引导与基于奖励的模型自适应关联起来。从带路径空间KL惩罚的终端奖励最大化出发,吉布斯变分原理与杜布变换生成对生成动力学的校正。局部去噪近似揭示了将输出偏好传输至受控变量的拉回映射。该构造在状态空间中得到变分最优引导(VOG),并通过将局部改进的端点摊销至模型参数,扩展为蒸馏变分最优控制(DVOC)。所得控制遵循局部变分规则:VOG通过去噪雅可比作用,DVOC则通过端点蒸馏实现参数拉回。在线性MMSE模型下,VOG的精确校正位于数据子空间。在ImageNet上,VOG在250步采样时将DiT-XL/2的FID从2.27降至2.04,在25步采样时从7.58降至4.44。在Z-Image-Turbo上经20次更新后,DVOC较未训练模型提升了HPSv3++及四个保留评估器的分数,在对比的训练方法中对应均值最高。这些结果通过共同几何原理将推理时控制与奖励自适应关联起来。

英文摘要:

We present a variational control framework that connects sampling-time guidance and reward-based model adaptation through a shared Jacobian pullback. Starting from terminal reward maximization with a path-space KL penalty, the Gibbs variational principle and Doob transform yield a correction to the generative dynamics. A local denoising approximation exposes the pullback that transports output preferences to the controlled variable. This construction gives Variational Optimal Guidance (VOG) in state space and extends to Distilled Variational Optimal Control (DVOC) by amortizing locally improved endpoints into model parameters. The resulting controls share a local variational rule: VOG acts through the denoising Jacobian, while DVOC realizes the parameter pullback by endpoint distillation. Under a linear MMSE model, VOG's exact correction lies in the data subspace. On ImageNet, VOG reduces DiT-XL/2 FID from 2.27 to 2.04 at 250 sampling steps and from 7.58 to 4.44 at 25 steps. After 20 updates on Z-Image-Turbo, DVOC improves HPSv3++ and four held-out evaluator scores over the untrained model, with the highest corresponding means among the trained methods compared. These results connect inference-time control and reward adaptation through a common geometric principle.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑