Inference-Time Compute Scaling For Flow Matching
This work addresses the under-explored area of inference-time compute scaling for FM, enabling applications in scientific domains like protein generation.
The paper tackles the problem of improving sample quality in Flow Matching (FM) by allocating extra computation at inference time, introducing novel scaling procedures that preserve the linear interpolant. Results show consistent quality improvements with increased compute, demonstrated on image and protein generation tasks.
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.