Research map / Architectures and efficiency / Mixture of experts
Combination-of-Experts with Knowledge Sharing for Cross-Task Vehicle Routing Problems
Zikang Yu, Jinbiao Chen, Jiahai Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace node-level MoE with combination-of-experts plus knowledge sharing (ReLD-MoEL to CoEKS) | ReLD · ID VRP n=50 | avg gap: 1.902% -> 1.751% | Improved |
| replace node-level MoE with CoEKS | ReLD · OOD VRP n=50 | avg gap: 4.202% -> 3.432% | Improved |
| replace node-level MoE with CoEKS | ReLD · OOD VRP n=100 | avg gap: 6.735% -> 5.857% | Improved |
| CoEKS versus parameter-matched ReLD-MoEL with 5 experts | ReLD-MoEL-E5 · OOD VRP n=50 | avg gap: 4.406% -> 3.432% | Improved |
| CoEKS versus parameter-matched ReLD-MoEL with 5 experts | ReLD-MoEL-E5 · ID VRP n=50 | avg gap: 2.069% -> 1.751% | Improved |
| plug in new constraint expert (CoEKS+) instead of EAL | ReLD-MoEL-EAL · VRPMB fine-tuning | gap: 2.59% -> 1.61% | Improved |
| initialize new expert from shared modules (CoEKS+c) versus random init | CoEKS+ · VRPMB fine-tuning | gap: 1.61% -> 1.56% | Improved |
| fine-tune new expert instead of adapter layers (AL) | ReLD-MoEL-AL · VRPMB fine-tuning | gap: 17.26% -> 1.56% | Improved |
| add multi-view knowledge sharing (MDis and shared layer) (ablation) | CoEKS · ID and OOD VRP n=50 | avg gap | Improved |
| apply MDis at higher encoder layers too (all layers) (ablation) | CoEKS · ID VRP n=50 | avg gap | Hurt |
| apply MDis only in lower layers versus deeper layers (ablation) | CoEKS · OOD VRP n=50 | avg gap | Mixed |
| increase distillation strength alpha beyond threshold (ablation) | CoEKS · ID and OOD VRP | avg gap | No clear effect |
Closest papers by meaning
- CaDA: Cross-Problem Routing Solver with Constraint-Aware Dual-Attention (ICML 2025 · similarity 0.90)
- Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under State-Decomposable MDP (NeurIPS 2025 · similarity 0.90)
- MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts (ICML 2024 · similarity 0.89)
- Rethinking Neural Combinatorial Optimization for Vehicle Routing Problems with Different Constraint Tightness Degrees (NeurIPS 2025 · similarity 0.88)
- Improving Generalization of Neural Combinatorial Optimization for Vehicle Routing Problems via Test-Time Projection Learning (NeurIPS 2025 · similarity 0.87)
- Collaboration! Towards Robust Neural Methods for Routing Problems (NeurIPS 2024 · similarity 0.87)
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.
