← 研踪 PaperTrailarXiv:2610.35607 · v1
VaLoP: Value Gradient Lookahead and Pullback for Reward-Guided Diffusion
首次提交 2026-10-26 23:29
版本更新 2026-10-26 23:29
雷达观测 2026-10-29 00:35
摘要
Reward guidance steers a pretrained diffusion or flow model toward images or actions that a reward model prefers, at sampling time and without retraining the model for each reward. It follows the gradient of a value function, the expected reward given the noisy state. The two common estimates of this gradient, gradients through a one-step denoised estimate and learned value models, are least accurate at high noise, where the coarse structure of the sample is decided. Interestingly, we find that both errors are posterior covariances, which grow with the uncertainty about the clean sample, and hence at higher noise. To reduce them, we evaluate the value at a less noisy state, one sampler step ahead, where this uncertainty is smaller, and pull its gradient back to the current state. This lookahead places both estimators at the two ends of one family and reveals that an intermediate lookahead requires one additional term, a correction through the generator. We therefore propose Value Gradient Lookahead and Pullback (VaLoP), which computes this correction with one vector-Jacobian product, and VaLoP-Fast, which amortizes it with a small head that, for text-to-image models, is trained with a single reward and reused for the others. Across various models, rewards, and benchmark datasets, both robustly outperform existing value-gradient guidance on the target reward, VaLoP-Fast at about the cost of learned-value guidance. Both extend to critic-guided robot policies, where VaLoP-Fast attains the highest success rates on LIBERO and RoboCasa.
作者备注：35 pages. Project page: https://real-lab-gist.github.io/valop
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v1 · 2026-10-26 23:29 · 最新观测版本
VaLoP: Value Gradient Lookahead and Pullback for Reward-Guided Diffusion
Reward guidance steers a pretrained diffusion or flow model toward images or actions that a reward model prefers, at sampling time and without retraining the model for each reward. It follows the gradient of a value function, the expected reward given the noisy state. The two common estimates of this gradient, gradients through a one-step denoised estimate and learned value models, are least accurate at high noise, where the coarse structure of the sample is decided. Interestingly, we find that both errors are posterior covariances, which grow with the uncertainty about the clean sample, and hence at higher noise. To reduce them, we evaluate the value at a less noisy state, one sampler step ahead, where this uncertainty is smaller, and pull its gradient back to the current state. This lookahead places both estimators at the two ends of one family and reveals that an intermediate lookahead requires one additional term, a correction through the generator. We therefore propose Value Gradient Lookahead and Pullback (VaLoP), which computes this correction with one vector-Jacobian product, and VaLoP-Fast, which amortizes it with a small head that, for text-to-image models, is trained with a single reward and reused for the others. Across various models, rewards, and benchmark datasets, both robustly outperform existing value-gradient guidance on the target reward, VaLoP-Fast at about the cost of learned-value guidance. Both extend to critic-guided robot policies, where VaLoP-Fast attains the highest success rates on LIBERO and RoboCasa.
查看此版本解读 →查看此版本原文 ↗