AI Paper Atlas

Research map / Learning theory and algorithms / Combinatorial and search optimization

Towards Efficient Constraint Handling in Neural Solvers for Routing Problems

Jieyi Bi, Zhiguang Cao, Jianan Zhou, Wen Song, Yaoxin Wu, Jie Zhang, Yining Ma, Cathy Wu

ICLR 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Unsupervised neural combinatorial optimization · 2 citations (OpenAlex)

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
add diversification loss L_DIV (ablation)CaR · TSPTWoptimality gap: 0.421% -> 0.014%Improved
add diversification loss L_DIV (ablation)CaR · CVRPoptimality gap: 0.325% -> 0.259%Improved
add supervised loss L_SL to weak backbone (ablation)CaR-POMO · TSPTWoptimality gap: 0.136% -> 0.014%Improved
add supervised loss L_SL to strong backbone (ablation)CaR-PIP · TSPTWoptimality gap: 0.006% -> 0.005%No clear effect
add supervised loss L_SL to weak backbone (ablation)CaR-POMO · TSPTWinfeasible ratio: 0.29% -> 0.01%Improved
add shared representation (unified encoder) (ablation)CaR · TSPTW-50 Hardoptimality gap: 0.199% -> 0.014%Improved
add shared representation (unified encoder) (ablation)CaR · TSPTW-100 Mediumoptimality gap: 7.589% -> 5.815%Improved
remove strict feasibility masking and use relaxed CMDPPOMO · CVRPBLTW-100optimality gap: 7.004% -> -1.645%Improved
use R&R instead of k-opt refinementCaR · CVRPBLTW-100optimality gap: -1.724% -> -2.448%Improved
add CaR joint training over naive construction+improvement combination (ablation)PIP · TSPTW-100optimality gap: 0.172% -> 0.146%Improved
train CaR on loose constraints instead of PIP on tight, test on tightCaR-PIP · TSPTW-100 Hardinfeasible ratio: 4.62% -> 3.52%Improved
add CaR refinement to PIP in cross-scale generalizationCaR · CVRPBLTW-200optimality gap: 4.51% -> 2.09%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