Unverified50% confidenceSolutionExact time
Reward guidance during diffusion model sampling is usually implemented by computing the gradient of the reward function with respect to the current sample and adding it, with a weight, to the denoising update, similar to classifier guidance or classifier-free guidance, but with the guidance signal coming from an external reward model.
1
Sources
50%
Confidence
Long-term
Relevance
10/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe denoising effect of the reward model propagates to the behavior of the final policy model, and a less noisy reward signal makes policy optimization more consistent with true human preferences.80% similarUnverifiedIn diffusion models, the score function is used to describe the log gradient of the data distribution and is trained via denoising score matching.77% similarUnverifiedDRaFT利用扩散模型去噪过程的可微性，将奖励函数梯度直接回传至模型参数，无需强化学习的策略梯度估计器，规避高方差问题76% similarVerifiedThe essence of reinforcement learning is to have the model continuously optimize to maximize a certain reward signal.74% similarVerifiedReward model training can use learning-to-rank frameworks such as the Bradley-Terry model, training a neural network to predict the probability of human preference between any two responses, taking a (prompt, response) pair as input and outputting a scalar score.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1080550API
curl https://kongchang.com/api/v1/knowledge/claims/1080550MCP
get_claim(id=1080550)