Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
CaDA: Cross-Problem Routing Solver with Constraint-Aware Dual-Attention
Authors: Han Li, Fei Liu, Zhi Zheng, Yu Zhang, Zhenkun Wang
ICML 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We comprehensively evaluate our model on 16 different VRPs and compare its performance against existing cross-problem VRP solvers. Ca DA achieves state-of-the-art results across all tested VRPs. Our ablation study confirms that each component contributes to its crossproblem learning performance. |
| Researcher Affiliation | Academia | 1Guangdong Provincial Key Laboratory of Fully Actuated System Control Theory and Technology, School of Automation and Intelligent Manufacturing, Southern University of Science and Technology, Shenzhen, China 2Department of Computer Science, City University of Hong Kong, Hong Kong, China. |
| Pseudocode | No | The paper describes the methodology using textual descriptions and mathematical formulations in Section 3 and Appendix B.3, but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code for Ca DA is publicly available at https:// github.com/CIAM-Group/Ca DA. |
| Open Datasets | Yes | We utilize the test dataset published by Routefinder (Berto et al., 2024b), which includes 1K randomly generated instances for each VRP variant, at scales of 50 and 100. ... To further validate the effectiveness of Ca DA in real-world instances, we conducted experiments using five test suites from CVRPLib1 benchmark datasets. These datasets comprise a total of 99 instances from Sets A, B, F, P, and X (Uchoa et al., 2017), with graph scales ranging from 16 to 200, various node distributions, and customer demands. 1http://vrp.atd-lab.inf.puc-rio.br/ |
| Dataset Splits | No | The paper mentions generating 1K randomly generated instances for testing each VRP variant (Section 4.2) and that models are trained on VRP50 and VRP100 (Section 4.1). Section C.1 details how problem instances are generated (e.g., node locations from uniform distribution, demands sampled), but it does not specify explicit training/validation/test dataset splits from a single, predefined dataset. It implies separate generation for training and testing. |
| Hardware Specification | Yes | All experiments are run on a platform with NVIDIA Ge Force RTX 3090 GPUs and Intel(R) Xeon(R) Gold 6348 CPUs at 2.60 GHz. |
| Software Dependencies | No | The paper mentions various algorithms and components like the REINFORCE algorithm, Transformer, RMSNorm, Swi GLU, and Adam W optimizer, but it does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | The hyperparameters used for training Ca DA are summarized in Table 5. (See table for details: Model Embedding dimension dh 128, Number of attention heads Mh 8, Number of encoder layers L 6, Top-k (N)/2, Feedforward hidden dimension da 512, Tanh clipping ΞΎ 10.0, Batch size 256, Train data per epoch 100,000, Optimizer Adam W, Learning rate (LR) 3e 4, Weight decay 1e 6, LR scheduler Multi Step LR, LR milestones [270, 295], LR gamma 0.1, Gradient clip value 1.0, Training epochs 300, Number of tasks used for training 16). |