AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

PoMtVRS: Preference-Optimized Multi-Task Vehicle Routing Solver with Preference Gating

Dian Meng, Yaoxin Wu, Yaqing Hou, Zhiguang Cao

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers

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 PGB and preference optimization to CaDACaDA · 16 VRPs, VRP50average gap: 1.721% -> 1.405%Improved
add PGB and preference optimization to CaDACaDA · 16 VRPs, VRP100average gap: 2.844% -> 2.205%Improved
add PGB and preference optimization to MVMoEMVMoE · 16 VRPs, VRP50/VRP100average gap: 2.227%/3.680% -> 1.809%/2.951%Improved
add preference-gated block (PGB) on top of PO (ablation)PoMtVRS(CaDA) · 16 VRPs, VRP50average gap: 1.579% -> 1.405%Improved
add head-wise gating in PGB (ablation)PoMtVRS(CaDA) · 16 VRPs, VRP100average gap: 2.241% -> 2.205%Improved
add state attribute mapping in PGB (ablation)PoMtVRS(CaDA) · 16 VRPs, VRP100average gap: 2.323% -> 2.205%Improved
train with preference optimization instead of RL (REINFORCE) (ablation)PoMtVRS(CaDA) · 16 VRPs, VRP50average gap: 1.621% -> 1.405%Improved
PO with ReLD-style decoder instead of RL (ablation)ReLD configuration · 16 VRPs, VRP50average gap: 1.714% -> 1.438%Improved
PoMtVRS(CaDA) instead of MoSES(CaDA)CaDA · 16 VRPs, VRP100average gap: 2.631% -> 2.205%Improved
add PoMtVRS to CaDACaDA · CVRPLIB Set-Xaverage gap: 9.230% -> 8.326%Improved
add PoMtVRS to CaDACaDA · Zero-shot MD, VRP50augmented gap: 37.571% -> 34.807%Improved
add efficient active search (EAS) to PoMtVRS(CaDA)PoMtVRS(CaDA) · 16 VRPs, 1K instancessolution qualityImproved

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