Navigating the Exploration-Exploitation Tradeoff in Inference-Time Scaling of Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Su, Xun, Huang, Jianming, Yusen, Yang, Fang, Zhongxi, Kasai, Hiroyuki
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908492741214208
author Su, Xun
Huang, Jianming
Yusen, Yang
Fang, Zhongxi
Kasai, Hiroyuki
author_facet Su, Xun
Huang, Jianming
Yusen, Yang
Fang, Zhongxi
Kasai, Hiroyuki
contents Inference-time scaling has achieved remarkable success in language models, yet its adaptation to diffusion models remains underexplored. We observe that the efficacy of recent Sequential Monte Carlo (SMC)-based methods largely stems from globally fitting the The reward-tilted distribution, which inherently preserves diversity during multi-modal search. However, current applications of SMC to diffusion models face a fundamental dilemma: early-stage noise samples offer high potential for improvement but are difficult to evaluate accurately, whereas late-stage samples can be reliably assessed but are largely irreversible. To address this exploration-exploitation trade-off, we approach the problem from the perspective of the search algorithm and propose two strategies: Funnel Schedule and Adaptive Temperature. These simple yet effective methods are tailored to the unique generation dynamics and phase-transition behavior of diffusion models. By progressively reducing the number of maintained particles and down-weighting the influence of early-stage rewards, our methods significantly enhance sample quality without increasing the total number of Noise Function Evaluations. Experimental results on multiple benchmarks and state-of-the-art text-to-image diffusion models demonstrate that our approach outperforms previous baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2508_12361
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Navigating the Exploration-Exploitation Tradeoff in Inference-Time Scaling of Diffusion Models
Su, Xun
Huang, Jianming
Yusen, Yang
Fang, Zhongxi
Kasai, Hiroyuki
Machine Learning
Artificial Intelligence
Statistics Theory
Inference-time scaling has achieved remarkable success in language models, yet its adaptation to diffusion models remains underexplored. We observe that the efficacy of recent Sequential Monte Carlo (SMC)-based methods largely stems from globally fitting the The reward-tilted distribution, which inherently preserves diversity during multi-modal search. However, current applications of SMC to diffusion models face a fundamental dilemma: early-stage noise samples offer high potential for improvement but are difficult to evaluate accurately, whereas late-stage samples can be reliably assessed but are largely irreversible. To address this exploration-exploitation trade-off, we approach the problem from the perspective of the search algorithm and propose two strategies: Funnel Schedule and Adaptive Temperature. These simple yet effective methods are tailored to the unique generation dynamics and phase-transition behavior of diffusion models. By progressively reducing the number of maintained particles and down-weighting the influence of early-stage rewards, our methods significantly enhance sample quality without increasing the total number of Noise Function Evaluations. Experimental results on multiple benchmarks and state-of-the-art text-to-image diffusion models demonstrate that our approach outperforms previous baselines.
title Navigating the Exploration-Exploitation Tradeoff in Inference-Time Scaling of Diffusion Models
topic Machine Learning
Artificial Intelligence
Statistics Theory
url https://arxiv.org/abs/2508.12361