Research map / Continual learning / Class-incremental
Lifelong Learning with Behavior Consolidation for Vehicle Routing
Jiyuan Pei, Yi Mei, Jialin Liu, Mengjie Zhang, Xin Yao
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 |
|---|---|---|---|
| add CaEW (confidence-aware experience weighting) instead of equal weights (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 5.2 -> 4.9 | Improved |
| use decision-seeking behavior consolidation instead of KLD (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 5.5 -> 4.9 | Improved |
| buffer experiences only at the last epoch of each task instead of every epoch (ablation) | LLR-BC · CVRP task order 1 | AF (x1e-3): 3.1 -> 0.6 | Improved |
| buffer one step as an experience instead of all steps of an instance (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 35.4 -> 4.9 | Improved |
| use variance instead of entropy for decision confidence (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 4.8 -> 4.9 | No clear effect |
| use variance instead of top-2 margin for decision confidence (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 4.8 -> 4.9 | No clear effect |
| keep original reservoir sampling instead of rescaled probabilities (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 4.9 -> 4.9 | No clear effect |
| set alpha to 10 instead of default (ablation) | LLR-BC · CVRP task order 1 | AP (x1e-3): 4.9 -> 13.2 | Hurt |
| use LLR-BC instead of fine-tuning with Omni base solver | Omni · CVRP lifelong | AP (x1e-3): 34.7 -> 16.5 | Improved |
| use LLR-BC instead of fine-tuning with INViT base solver | INViT · TSP lifelong | AF (x1e-3): 0.6 -> 4.5 | Hurt |
| use LLR-BC instead of fine-tuning | Fine-tuning · CVRP lifelong (mean over 5 task orders) | AP (x1e-3): 23.5 -> 4.2 | Improved |
| use LLR-BC instead of EWC | EWC · TSP lifelong (mean over 5 task orders) | AP (x1e-3): 18.3 -> 3.4 | Improved |
Closest papers by meaning
- Memory-Enhanced Neural Solvers for Routing Problems (NeurIPS 2025 · similarity 0.88)
- Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under State-Decomposable MDP (NeurIPS 2025 · similarity 0.87)
- A Mixed-Curvature based Pre-training Paradigm for Multi-Task Vehicle Routing Solver (ICML 2025 · similarity 0.87)
- MTL-KD: Multi-Task Learning Via Knowledge Distillation for Generalizable Neural Vehicle Routing Solver (NeurIPS 2025 · similarity 0.87)
- MVMoE: Multi-Task Vehicle Routing Solver with Mixture-of-Experts (ICML 2024 · similarity 0.86)
- Neural Solver Selection for Combinatorial Optimization (ICML 2025 · similarity 0.86)
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.
