在推理时分配额外计算,最近已提升了大语言模型和基于扩散的图像生成中的样本质量。与此同时,Flow Matching(FM,流匹配)已在语言、视觉和科学领域受到关注,但其推理时扩展方法仍未得到充分研究。同时,Kim 等人(2025)处理了这一问题,但在推理时用非线性方差保持(VP)插值替换了线性插值,牺牲了 FM 高效且笔直的采样。此外,Flow Matching 的推理时计算扩展此前仅应用于图像生成等视觉任务。我们提出了用于 FM 的新型推理时扩展流程,在采样期间保留线性插值。在图像生成以及(据我们所知首次)无条件蛋白质生成上对我们的方法进行评估,结果表明:I)样本质量随推理计算增加而持续提升;II)Flow Matching 推理时扩展可应用于科学领域。
英文摘要:
Allocating extra computation at inference time has recently improved sample quality in large language models and diffusion-based image generation. In parallel, Flow Matching (FM) has gained traction in language, vision, and scientific domains, but inference-time scaling methods for it remain under-explored. Concurrently, Kim et al., 2025 approach this problem but replace the linear interpolant with a non-linear variance-preserving (VP) interpolant at inference, sacrificing FM's efficient and straight sampling. Additionally, inference-time compute scaling for flow matching has only been applied to visual tasks, like image generation. We introduce novel inference-time scaling procedures for FM that preserve the linear interpolant during sampling. Evaluations of our method on image generation, and for the first time (to the best of our knowledge), unconditional protein generation, show that I) sample quality consistently improves as inference compute increases, and II) flow matching inference-time scaling can be applied to scientific domains.
Scaling Clinical Trial Matching Using Large Language Models: A Case Study in Oncology
Microsoft Research(微软研究院);Providence Health & Services(普罗维登斯医疗服务机构);Earle A. Chiles Research Institute(厄尔·A·奇尔斯研究所);Providence Cancer Institute(普罗维登斯癌症研究所)
Graduate School of Advanced Science and Engineering Hiroshima University(广岛大学先进科学与工程学院);Intelligent Production Technology Research & Development Center for Aerospace Gifu University(岐阜大学航空航天智能生产技术研究开发中心)