AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Native Adaptive Solution Expansion for Diffusion-based Combinatorial Optimization

Yu Wang, Yang Li, Jiale Ma, Junchi Yan, Yi Chang

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Distillation for few-step generation

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 CO-specific one-way mask corruption instead of uniform bit-flip corruption (ablation)NExCO · TSP-500gap: 0.59% -> 0.28% (5 steps)Improved
use CO-specific one-way mask corruption instead of uniform bit-flip corruption (ablation)NExCO · TSP-1000gap: 0.86% -> 0.63% (5 steps)Improved
use native adaptive expansion instead of non-adaptive global t-schedule (ablation)NExCO · TSP-500gap: 0.71% -> 0.28% (5 steps)Improved
use native adaptive expansion instead of non-adaptive global t-schedule (ablation)NExCO · TSP-1000gap: 0.96% -> 0.63% (5 steps)Improved
use feasibility projection instead of greedy decoding with remasking (ablation)NExCO · TSP-500time: 0.54s -> 0.33s at similar gap (0.31% vs 0.28%)Improved
train on 2-Opt perturbed suboptimal labels (ablation)NExCO · TSP-500gap: 0.25% -> 0.26% (Ds = 7)No clear effect
increase expansion steps Ds from 3 to 7 (ablation)NExCO · TSP-500gap: 0.39% -> 0.25%Improved
set candidate threshold alpha too high or too low (ablation)NExCO · TSP, MIS, CVRPsolution qualityHurt
replace Fast T2T with NExCO as decoderdiffusion CO solver · MIS ER-[700-800]drop vs KaMIS: 9.31% -> 4.20%Improved
train on larger TSP instances before cross-scale testing (ablation)NExCO · TSP-1000 testgap: 3.86% -> 0.58% (train TSP-100 -> TSP-500)Improved

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