{"id":"arxiv/2605.17850","url":"https://arxiv.org/abs/2605.17850","tags":["stat.ML","cs.CV","cs.LG","math.NA","math.PR"],"score":0,"title":"Simple Approximation and Derivative Free Inference-Time Scaling for Diffusion Models via Sequential Monte Carlo on Path Measures","domain":"stat.ML","source":"arxiv","authors":["Chenyang Wang","Weizhong Wang","Yinuo Ren","Jose Blanchet","Yiping Lu"],"reviews":0,"subject":"stat.ML","summary":"iffusion-based generative models increasingly rely on inference-time guidance, adding a drift term or reweighting mixture of experts, to improve sample quality on task-specific objectives. However, most existing techniques require repeated score or gradient evaluations, introducing bias, high computational overhead, or both. We introduce \\texttt{URGE}, Unbiased Resampling via Girsanov Estimation, a derivative-free inference-time scaling algorithm that performs path-wise importance reweighting via a Girsanov change of measure. Instead of computing gradient-based particle weights in previous…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-05-18","reactions":{}}