AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Inference-time scaling of diffusion models through classical search

XiangCheng Zhang, Haowei Lin, Haotian Ye, James Zou, Jianzhu Ma, Yitao Liang, Yilun Du

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Training-free diffusion acceleration

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use SSP resampling instead of multinomial (ablation)BFS tree search · SD v1.5 ImageReward, N=4ImageReward score: 0.743 -> 0.834Improved
Use Increase tempering instead of constant (ablation)BFS tree search · SD v1.5 ImageReward, N=4ImageReward score: 0.834 -> 0.882Improved
Use infinite tempering (top particle only) instead of constant (ablation)BFS tree search · SD v1.5 ImageReward, N=4ImageReward score: 0.834 -> 0.667Hurt
Use Max scoring instead of Current (ablation)BFS tree search · SD v1.5 ImageReward, N=4ImageReward score: 0.812 -> 0.834Improved
Use improved BFS instead of prior particle-based methodsBFS · SDXL ImageReward, N=8ImageReward score: DAS 1.265 -> 1.291Improved
Replace BFS with adaptive DFS with backtrackingDFS · CompBench T2Iaccuracy vs computeImproved
Scale local Langevin search steps together with global searchBest-of-N / BFS / DFS · PointMaze planningsuccess vs computeImproved
Use local search only (TFG) without global searchTFG · PointMaze planningsuccess vs computeHurt
Use test-time search with local and global search instead of TFG and DASTTS · D4RL Medium Hoppernormalized score: DAS 71.3 -> 99.5Improved
Compare training-free TTS with training-based QGPOTTS · D4RL locomotion averagenormalized score: QGPO 86.6 vs TTS 86.1No clear effect
Distill TTS actions into diffusion policyTTS-distill · D4RL Medium Hoppernormalized score: DPPO 92.8 -> 98.8Improved
Use separate verifiers for local and global search (double verifier) (ablation)BFS · ImageNet conditional, N=8FID: 133.3 -> 118.2Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy