Latent Adversarial Regularization for Offline Preference Optimization
This work addresses offline preference optimization for language models, offering an incremental improvement over token-level regularization methods.
The paper tackled the challenge of preference optimization for language models by introducing latent adversarial regularization to penalize divergence between internal representations of policy and reference models, resulting in more robust structural feedback under distributional shift and noise with comparable downstream performance and minor computational overhead.
Learning from human feedback typically relies on preference optimization that constrains policy updates through token-level regularization. However, preference optimization for language models is particularly challenging because token-space similarity does not imply semantic or behavioral similarity. To address this challenge, we leverage latent-space regularization for language model preference optimization. We introduce GANPO, which achieves latent-space regularization by penalizing divergence between the internal representations of a policy model and a reference model. Given that latent representations are not associated with explicit probability densities, we adopt an adversarial approach inspired by GANs to minimize latent-space divergence. We integrate GANPO as a regularizer into existing offline preference optimization objectives. Experiments across multiple model architectures and tasks show consistent improvements from latent-space regularization. Further, by comparing GANPO-induced inferential biases with those from token-level regularization, we find that GANPO provides more robust structural feedback under distributional shift and noise while maintaining comparable downstream performance with minor computational overhead.