无需重训练即可高效控制离散扩散模型生成，提升序列设计性能。
Plug-and-Play Guidance for Discrete Diffusion Models via Gradient-Informed Logit Correction
- 用预训练去噪网络做变分代理，间接估计梯度信号。
- 直接修正干净预测的逻辑值，避免高维离散空间梯度不稳。
- 支持可微与不可微奖励函数，适合生物序列生成任务。
可控生成在离散扩散模型中常受计算开销大或需重训练的限制。本文提出梯度感知逻辑值修正（GILC），一种即插即用框架，通过复用预训练去噪网络作为变分代理，高效估算引导信号。为克服高维离散空间固有的梯度不稳定性，引入无雅可比机制，直接修正干净预测的逻辑值，实现稳定有效的引导。该方法兼容可微与不可微奖励函数。在DNA、蛋白质序列和分子生成任务上的大量实验表明，GILC无需额外训练即达领先性能，通常优于微调方法。
原文摘要 · Abstract (English)
Controllable generation with discrete diffusion models is often hindered by high computational overhead or the need for retraining. In this paper, we present \underline{\textbf{G}}radient-\underline{\textbf{I}}nformed \underline{\textbf{L}}ogit \underline{\textbf{C}}orrection (\textbf{GILC}), a plug-and-play framework that efficiently estimates guidance signals by repurposing the pretrained denoising network as a variational proxy. To circumvent the gradient instability inherent in high-dimensional discrete spaces, we introduce a Jacobian-free mechanism that directly corrects the clean prediction logits, facilitating stable and effective guidance. Our method accommodates both differentiable and non-differentiable reward functions. Extensive experiments across DNA, protein sequence, and molecular generation tasks demonstrate that GILC achieves state-of-the-art performance without additional training, frequently outperforming fine-tuning approaches.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。