Abstract:• A data structure is proposed to store the resolving relations of graphs, on which the greedy repair policy for solving the MDP is developed. • A hybrid approximate algorithm for solving the MDP is established, which combines graph neural networks and greedy repair policy. By designing specific reward function related to the background of the MDP, the policy gradient algorithm is used to learn the extraction of graph metric base. • Complex networks and random graphs are established, on which the advantages and disadvantages of different graph neural network architectures in learning the metric base of graphs are compared. Moreover, the computational accuracy of graph neural networks and the effect of greedy repair policy are analyzed respectively. Furthermore, the convergence of graph neural networks, the characteristics of the data structure and the complexity of the hybrid approximate algorithm are all considered. For distinct vertices x , y , z in graph G = ( V , E ), we say the pair of vertices x , y can be distinguished by vertex z, if the distance inequation d G ( x , z ) ≠ d G ( y , z ) is true. If any two different vertices in graph G can be distinguished by at least one vertex in the vertex subset S ⊂ V then we say S ⊂ V is a resolving set of graph G. The one with minimum number of nodes among all of the resolving sets is called a metric base of graph G, of which the cardinality is called the metric dimension of graph G. The metric dimension problem(MDP) of computing the metric dimension of graphs is a kind of complex combinatorial optimization problem. In this paper, a hybrid algorithm for solving the MDP is developed, through which the metric base of graphs can be learned, and consequently the metric dimension of graphs is estimated. It is particular that the hybrid algorithm combines graph representation learning with greedy repair policy. Extensive experiments show that the hybrid algorithm has higher computational accuracy and efficiency on random graphs. It is also further suggested that the learning process of MDP is greatly influenced by the attributes of complex networks and the designs of graph neural networks themselves.

Graph Metric Learning Via Gershgorin Disc Alignment

Signed Graph Metric Learning via Gershgorin Disc Perfect Alignment

Projection-free Graph-based Classifier Learning Using Gershgorin Disc Perfect Alignment

Metric Learning Via Penalized Optimization.

Scalable unsupervised alignment of general metric and non-metric structures

Data-Adaptive Metric Learning with Scale Alignment.

Fast Graph Sampling Set Selection Using Gershgorin Disc Alignment

Deep Metric Learning with Graph Consistency

Efficient Directed Graph Sampling via Gershgorin Disc Alignment

Geometry-Aware Graph Embedding Projection Metric Learning for Image Set Classification

Reconstruction-Cognizant Graph Sampling Using Gershgorin Disc Alignment

Toward the Optimization of Normalized Graph Laplacian

Towards Generalized and Efficient Metric Learning on Riemannian Manifold.

Online Low-Rank Metric Learning via Parallel Coordinate Descent Method

Robust Metric Learning on Grassmann Manifolds with Generalization Guarantees.

Fast Computation of Generalized Eigenvectors for Manifold Graph Embedding

A Novel Metric Learning Framework by Exploiting Global and Local Information

Learning to Compute the Metric Dimension of Graphs

Scalable Deep Metric Learning on Attributed Graphs

Graph Representation Learning with Adaptive Metric

Feature Graph Learning for 3D Point Cloud Denoising