Abstract:Travelling Salesperson Problems (TSPs) and Vehicle Routing Problems (VRPs) have achieved reasonable improvement in accuracy and computation time with the adaptation of Machine Learning (ML) methods. However, none of the previous works completely respects the symmetries arising from TSPs and VRPs including rotation, translation, permutation, and scaling. In this work, we introduce the first-ever completely equivariant model and training to solve combinatorial problems. Furthermore, it is essential to capture the multiscale structure (i.e. from local to global information) of the input graph, especially for the cases of large and long-range graphs, while previous methods are limited to extracting only local information that can lead to a local or sub-optimal solution. To tackle the above limitation, we propose a Multiresolution scheme in combination with Equivariant Graph Attention network (mEGAT) architecture, which can learn the optimal route based on low-level and high-level graph resolutions in an efficient way. In particular, our approach constructs a hierarchy of coarse-graining graphs from the input graph, in which we try to solve the routing problems on simple low-level graphs first, then utilize that knowledge for the more complex high-level graphs. Experimentally, we have shown that our model outperforms existing baselines and proved that symmetry preservation and multiresolution are important recipes for solving combinatorial problems in a data-driven manner. Our source code is publicly available at <a class="link-external link-https" href="https://github.com/HySonLab/Multires-NP-hard" rel="external noopener nofollow">this https URL</a>
What problem does this paper attempt to address?
This paper attempts to address two major challenges in the Traveling Salesman Problem (TSP) and the Vehicle Routing Problem (VRP):
1. **Invariance to geometric transformations**: TSP and VRP are represented in two - dimensional Euclidean space, and their solutions must be invariant to geometric transformations (such as rotation, permutation, translation, and scaling) of the input city coordinates. However, most existing machine - learning - based methods, such as the Attention Model, have not fully considered the underlying symmetries of routing problems.
2. **Learning of multi - scale information**: Current learning - based methods usually focus on learning from local information (i.e., spatial locality) to improve the quality of solutions, which may cause the model to converge to a local optimum and ignore the importance of global information. For large and long - range graphs, relying solely on local information may lead to sub - optimal solutions.
To solve the above problems, the authors propose a multi - resolution equivariant graph attention network (mEGAT) architecture. The specific methods are as follows:
- **Equivariant graph attention model**: A fully equivariant model and training method are introduced to solve combinatorial optimization problems and ensure that the model is more stable when dealing with combinatorial optimization problems with invariant transformations.
- **Multi - resolution learning framework**: By constructing a hierarchical coarse - grained graph, the problem is first solved on a simple low - level graph, and then this knowledge is used to solve more complex high - level graphs. This method can capture both local and global information of the input graph simultaneously, thus avoiding getting trapped in local optimal solutions.
### Main contributions of the paper
1. **Identification and application of equivariance and symmetry**: For the first time, the equivariance and symmetry that need to be considered when solving routing problems in deep (reinforcement) models are identified and proposed, and it is suggested to use the equivariant graph attention model to respect the invariant transformations of the input graph.
2. **Multi - resolution graph training**: A multi - resolution graph training method is proposed, which can learn routing problems at multiple levels (sub - graphs and high - level graphs). The model quickly captures the local and implicit global structures of the problem by sharing weights of original problem instances at different scales.
3. **Experimental verification**: The effectiveness and efficiency of the model are verified on large synthetic and real - world datasets, demonstrating the superiority and stability of this model in solution quality, especially when dealing with TSP problems with various distributions, sizes, and symmetries.
Through these improvements, the authors hope to significantly reduce the computation time without sacrificing the quality of solutions and make the model more generalizable when solving other combinatorial optimization problems.