DrPO：一步生成模型偏好优化的新方法
Drifting Preference Optimization for One-Step Generative Models
DrPO 解决了单步生成模型偏好微调的核心痛点——无需可微奖励或复杂去噪轨迹，做文本到图像生成的团队可以直接用黑箱奖励提升模型对齐度，训练效率还提升了3倍多，值得关注。
一步式文本到图像生成器（如SD-Turbo）因单次前向传播即可生成图像而备受关注，但其偏好微调面临挑战。现有方法依赖策略似然、去噪轨迹或可微奖励梯度，难以直接应用。
研究者提出Drifting Preference Optimization (DrPO)，一种在线偏好微调方法，通过从当前生成器采样候选图像，用目标奖励排序，并合成特征空间更新方向（非参数偶极偏好场加参考漂移），实现无需奖励梯度的训练。
DrPO在SD-Turbo和SDXL-Turbo上评估，使用HPSv3和GenEval等基准，相比无奖励梯度的一步偏好基线提升了对齐效果，并在匹配有效批次设置下将HPSv3训练计算量降低3.51倍。
该方法支持大型、黑箱或不可微奖励，且推理时仍保持单次生成调用。
Drifting Preference Optimization for One-Step Generative Models
One-step text-to-image generators are attractive for deployment because they generate an image with a single forward pass, but preference finetuning them remains difficult: standard alignment methods often rely on policy likelihoods, denoising trajectories, differentiable reward gradients, or test-time optimization. We propose Drifting Preference Optimization (DrPO), an online preference-finetuning method for deterministic one-step generators. For each prompt, DrPO samples candidates from the current generator, ranks them with a target reward, and uses high- and low-scoring samples to synthesize a feature-space update direction. The update is a non-parametric dipole preference field plus a reference drift estimated from the frozen base generator, and is optimized through a detached feature-space regression target. The target reward is used only for ranking, so DrPO can train with large, black-box, or non-differentiable rewards while inference remains a single generator call. We evaluate DrPO on SD-Turbo and SDXL-Turbo with multiple target rewards and benchmarks, including HPSv3 and GenEval. DrPO improves alignment over reward-gradient-free one-step preference baselines and reduces HPSv3 training computation by $3.51\times$ under the matched effective-batch setting by removing reward-model backpropagation. Initial offline experiments suggest that sample-based gradient synthesis can also be used beyond online reward ranking.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include preference optimization, text-to-image, SD-Turbo, reward function.
Source attribution: curated from arXiv cs.LG (@Zhou Jiang, Yandong Wen, Zhen Liu).