Learning from Noisy Preferences: A Semi-Supervised Learning Approach to Direct Preference Optimization
For practitioners aligning diffusion models with human preferences, this work tackles the problem of noisy binary labels from multi-dimensional preferences, offering a practical solution that improves alignment quality.
The paper addresses label noise in human preference datasets for diffusion model alignment, where multi-dimensional preferences are compressed into binary labels. Semi-DPO, a semi-supervised approach, achieves state-of-the-art performance and significantly improves alignment with complex human preferences without additional annotation or reward models.
Human visual preferences are inherently multi-dimensional, encompassing aesthetics, detail fidelity, and semantic alignment. However, existing datasets provide only single, holistic annotations, resulting in severe label noise: images that excel in some dimensions but are deficient in others are simply marked as winner or loser. We theoretically demonstrate that compressing multi-dimensional preferences into binary labels generates conflicting gradient signals that misguide Diffusion Direct Preference Optimization (DPO). To address this, we propose Semi-DPO, a semi-supervised approach that treats consistent pairs as clean labeled data and conflicting ones as noisy unlabeled data. Our method starts by training on a consensus-filtered clean subset, then uses this model as an implicit classifier to generate pseudo-labels for the noisy set for iterative refinement. Experimental results demonstrate that Semi-DPO achieves state-of-the-art performance and significantly improves alignment with complex human preferences, without requiring additional human annotation or explicit reward models during training. We will release our code and models at: https://github.com/L-CodingSpace/semi-dpo