Learning from Noisy Preferences: A Semi-Supervised Learning Approach to Direct Preference Optimization

TL;DR

Semi-DPO is a semi-supervised learning method that improves preference optimization by effectively handling noisy, multi-dimensional human preference data, achieving state-of-the-art results.

Contribution

It introduces a semi-supervised approach that distinguishes clean and noisy preference data, enhancing alignment with complex human preferences without extra annotations.

Findings

Semi-DPO outperforms existing methods in preference alignment.

It effectively filters and utilizes noisy preference data.

The approach achieves state-of-the-art performance on benchmark datasets.

Abstract

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