提出轻量级方法，让扩散模型推理时高效适配新分布。
DriftLite: Lightweight Drift Control for Inference-Time Scaling of Diffusion Models
- 利用福克-普朗克方程中漂移与势能的自由度，实时调整采样路径。
- 在多类任务中降低样本方差，生成质量优于传统引导和蒙特卡洛方法。
- 无需训练，适合需快速适配的扩散模型应用，如蛋白质折叠。
我们研究扩散模型的推理时缩放问题，目标是在不重新训练的前提下，将预训练模型适配到新目标分布。现有基于引导的方法虽简单但引入偏差，而基于粒子的修正方法则存在权重退化和高计算开销问题。我们提出 DriftLite，一种轻量、免训练的基于粒子的方法，可在线提供最优稳定性控制以引导推理动态。DriftLite 利用福克-普朗克方程中漂移与粒子势能间未被探索的自由度，实现了两种实用变体：方差控制引导（VCG）和能量控制引导（ECG），以最小开销逼近最优漂移。在高斯混合模型、粒子系统及大规模蛋白质-配体共折叠问题上，DriftLite 均持续降低方差并提升样本质量，优于纯引导法和序列蒙特卡洛基线。这些结果揭示了可扩展推理时适应扩散模型的一种原理性、高效路径。代码已公开于 https://github.com/yinuoren/DriftLite。
原文摘要 · Abstract (English)
We study inference-time scaling for diffusion models, where the goal is to adapt a pre-trained model to new target distributions without retraining. Existing guidance-based methods are simple but introduce bias, while particle-based corrections suffer from weight degeneracy and high computational cost. We introduce DriftLite, a lightweight, training-free particle-based approach that steers the inference dynamics on the fly with provably optimal stability control. DriftLite exploits a previously unexplored degree of freedom in the Fokker-Planck equation between the drift and particle potential, and yields two practical instantiations: Variance- and Energy-Controlling Guidance (VCG/ECG) for approximating the optimal drift with minimal overhead. Across Gaussian mixture models, particle systems, and large-scale protein-ligand co-folding problems, DriftLite consistently reduces variance and improves sample quality over pure guidance and sequential Monte Carlo baselines. These results highlight a principled, efficient route toward scalable inference-time adaptation of diffusion models. Our source code is publicly available at https://github.com/yinuoren/DriftLite.
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。