Understanding the Performance Gap in Preference Learning: A Dichotomy of RLHF and DPO
This provides theoretical insights for researchers and practitioners in AI alignment on when to choose RLHF or DPO, though it is incremental as it builds on existing methods.
The paper analyzes the performance gap between RLHF and DPO in preference learning, showing that online DPO can outperform both under certain model mis-specifications and that RLHF requires significantly fewer samples than DPO in sparse reward settings.
We present a fine-grained theoretical analysis of the performance gap between reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) under a representation gap. Our study decomposes this gap into two sources: an explicit representation gap under exact optimization and an implicit representation gap under finite samples. In the exact optimization setting, we characterize how the relative capacities of the reward and policy model classes influence the final policy qualities. We show that RLHF, DPO, or online DPO can outperform one another depending on type of model mis-specifications. Notably, online DPO can outperform both RLHF and standard DPO when the reward and policy model classes are isomorphic and both mis-specified. In the approximate optimization setting, we provide a concrete construction where the ground-truth reward is implicitly sparse and show that RLHF requires significantly fewer samples than DPO to recover an effective reward model, highlighting a statistical advantage of two-stage learning. Together, these results provide a comprehensive understanding of the performance gap between RLHF and DPO under various settings, and offer practical insights into when each method is preferred.