Towards Robust Alignment of Language Models: Distributionally Robustifying Direct Preference Optimization
This work addresses noise robustness in aligning language models with human preferences, which is an incremental improvement over existing DPO methods.
This study tackled noise in training datasets for Direct Preference Optimization (DPO) by categorizing noise into pointwise and pairwise types and using Distributionally Robust Optimization (DRO) to enhance DPO's resilience, resulting in Dr. DPO, which substantially improved text quality and response accuracy in both noisy and noise-free settings.
This study addresses the challenge of noise in training datasets for Direct Preference Optimization (DPO), a method for aligning Large Language Models (LLMs) with human preferences. We categorize noise into pointwise noise, which includes low-quality data points, and pairwise noise, which encompasses erroneous data pair associations that affect preference rankings. Utilizing Distributionally Robust Optimization (DRO), we enhance DPO's resilience to these types of noise. Our theoretical insights reveal that DPO inherently embeds DRO principles, conferring robustness to pointwise noise, with the regularization coefficient $β$ playing a critical role in its noise resistance. Extending this framework, we introduce Distributionally Robustifying DPO (Dr. DPO), which integrates pairwise robustness by optimizing against worst-case pairwise scenarios. The novel hyperparameter $β'$ in Dr. DPO allows for fine-tuned control over data pair reliability, providing a strategic balance between exploration and exploitation in noisy training environments. Empirical evaluations demonstrate that Dr. DPO substantially improves the quality of generated text and response accuracy in preference datasets, showcasing enhanced performance in both noisy and noise-free settings. The code is available at https://github.com/junkangwu/Dr_DPO.