HPS: Hard Preference Sampling for Human Preference Alignment
This addresses the challenge of building safe and controllable AI systems by improving preference alignment, though it appears incremental as it builds on existing Plackett-Luce and Bradley-Terry methods.
The paper tackles the problem of aligning Large Language Model responses with human preferences by proposing Hard Preference Sampling (HPS), a framework that prioritizes preferred responses while rejecting dispreferred and harmful ones, resulting in improved reward margins and reduced harmful content generation on HH-RLHF and PKU-Safety datasets.
Aligning Large Language Model (LLM) responses with human preferences is vital for building safe and controllable AI systems. While preference optimization methods based on Plackett-Luce (PL) and Bradley-Terry (BT) models have shown promise, they face challenges such as poor handling of harmful content, inefficient use of dispreferred responses, and, specifically for PL, high computational costs. To address these issues, we propose Hard Preference Sampling (HPS), a novel framework for robust and efficient human preference alignment. HPS introduces a training loss that prioritizes the most preferred response while rejecting all dispreferred and harmful ones. It emphasizes "hard" dispreferred responses -- those closely resembling preferred ones -- to enhance the model's rejection capabilities. By leveraging a single-sample Monte Carlo sampling strategy, HPS reduces computational overhead while maintaining alignment quality. Theoretically, HPS improves sample efficiency over existing PL methods and maximizes the reward margin between preferred and dispreferred responses, ensuring clearer distinctions. Experiments on HH-RLHF and PKU-Safety datasets validate HPS's effectiveness, achieving comparable BLEU and reward scores while greatly improving reward margins and thus reducing harmful content generation.