Abstract:Graphons are continuous models that represent the structure of graphs and allow the generation of graphs of varying sizes. We propose Scalable Implicit Graphon Learning (SIGL), a scalable method that combines implicit neural representations (INRs) and graph neural networks (GNNs) to estimate a graphon from observed graphs. Unlike existing methods, which face important limitations like fixed resolution and scalability issues, SIGL learns a continuous graphon at arbitrary resolutions. GNNs are used to determine the correct node ordering, improving graph alignment. Furthermore, we characterize the asymptotic consistency of our estimator, showing that more expressive INRs and GNNs lead to consistent estimators. We evaluate SIGL in synthetic and real-world graphs, showing that it outperforms existing methods and scales effectively to larger graphs, making it ideal for tasks like graph data augmentation.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to solve the problem of estimating latent graphon models from multiple (possibly large - scale) networks. Specifically, a graphon is a continuous model used to represent graph structures and can generate graphs of different scales. However, in practical applications, the true graphon model is usually unknown and needs to be estimated from the observed graphs.
#### Main challenges:
1. **Limitations of existing methods**: Existing graphon estimation methods have some important limitations, such as fixed - resolution and scalability issues. These methods usually treat the graphon as a fixed - resolution matrix, limiting the range of graphon functions they can capture. In addition, methods relying on predefined ranking metrics (such as node degree or k - core) may fail when dealing with nodes with similar characteristics.
2. **Graph alignment problem**: In order to accurately estimate the graphon, it is necessary to align the observed graphs. Traditional alignment methods rely on predefined metrics (such as node degree), which are not effective when nodes have similar characteristics.
3. **Processing of large - scale graphs**: As the graph scale increases, existing methods based on Gromov - Wasserstein (GW) distance become impractical because the combinatorial nature of their optimization problems leads to a sharp increase in computational complexity.
#### Proposed solutions:
The paper proposes a new method named Scalable Implicit Graphon Learning (SIGL), which combines implicit neural representation (INR) and graph neural network (GNN) to estimate the graphon. The main features of SIGL include:
- **Continuous graphon estimation at arbitrary resolution**: Unlike existing methods, SIGL can learn continuous graphon models at arbitrary resolutions, rather than being limited to a fixed resolution.
- **GNN - based node ranking**: By using GNN to learn the latent variables of nodes, the correct node ranking is estimated, improving the effect of graph alignment.
- **Efficient training framework**: SIGL uses the mean - squared error (MSE) as a loss function instead of the complex GW distance, making this method more efficient in processing large - scale graphs.
- **Asymptotic consistency analysis**: The paper also formally characterizes the asymptotic consistency of the SIGL estimator, proving that more expressive INR and GNN can lead to consistent estimators.
#### Experimental verification:
The authors verified the effectiveness and scalability of SIGL through experiments on synthetic graphs and real - world graphs. The experimental results show that SIGL outperforms existing methods in estimating graphon models and exhibits better performance and faster training speed on large - scale graphs.
Overall, this paper solves key problems in graphon estimation, especially the efficient processing of large - scale graphs and the graph alignment problem, providing new tools and methods for graph data analysis.