Abstract:In this paper, we present GRASP, a novel graph generative model based on 1) the spectral decomposition of the graph Laplacian matrix and 2) a diffusion process. Specifically, we propose to use a denoising model to sample eigenvectors and eigenvalues from which we can reconstruct the graph Laplacian and adjacency matrix. Our permutation invariant model can also handle node features by concatenating them to the eigenvectors of each node. Using the Laplacian spectrum allows us to naturally capture the structural characteristics of the graph and work directly in the node space while avoiding the quadratic complexity bottleneck that limits the applicability of other methods. This is achieved by truncating the spectrum, which as we show in our experiments results in a faster yet accurate generative process. An extensive set of experiments on both synthetic and real world graphs demonstrates the strengths of our model against state-of-the-art alternatives.

What problem does this paper attempt to address?

The main focus of this paper is graph generation, particularly how to generate graphs through spectral diffusion. Existing graph generation models have limitations in dealing with complex network structures and cannot directly learn from observed data. The paper proposes a new model called GRASP, which is based on spectral decomposition and diffusion process of the graph Laplacian matrix. GRASP uses a denoising model to sample feature vectors and eigenvalues, and then reconstructs the graph Laplacian and adjacency matrix. In this way, it can work directly in the node space and avoid computational complexity bottlenecks by truncating the spectrum, thus achieving faster and more accurate generation process. The advantage of GRASP is that it can directly generate eigenvalues and eigenvectors instead of interpolating from real data like other models. Moreover, it can naturally handle discrete and continuous node features. By working in the spectral domain, GRASP overcomes computational bottlenecks while capturing the structural characteristics of the graph. Experiments show that GRASP outperforms current state-of-the-art alternative methods on synthetic and real-world graph datasets, particularly in terms of expressive power and realism of the generated graphs. The paper also discusses the limitations of traditional graph generation methods such as the Barabási-Albert and Kronecker models, which fail to capture the complex dependencies in real-world networks adequately. In contrast, GRASP leverages advancements in deep learning techniques, particularly variational autoencoders, recurrent neural networks, and generative adversarial networks, to more effectively generate graphs with diverse structural characteristics. Although there are already related models in applications such as molecule generation, GRASP has advantages in dealing with large graphs and node features.

Graph Generation via Spectral Diffusion

Fast Graph Generation via Spectral Diffusion

Sparse Training of Discrete Diffusion Models for Graph Generation

Autoregressive Diffusion Model for Graph Generation

SaGess: Sampling Graph Denoising Diffusion Model for Scalable Graph Generation

Graphusion: Latent Diffusion for Graph Generation

Diffusion Models for Graphs Benefit From Discrete State Spaces

Graph Generation with Diffusion Mixture

GRASPEL: Graph Spectral Learning at Scale

GraphGDP: Generative Diffusion Processes for Permutation Invariant Graph Generation

Pard: Permutation-Invariant Autoregressive Diffusion for Graph Generation

Hyperbolic Geometric Latent Diffusion Model for Graph Generation

Generative Diffusion Models on Graphs: Methods and Applications

Efficient and Scalable Graph Generation through Iterative Local Expansion

inGRASS: Incremental Graph Spectral Sparsification via Low-Resistance-Diameter Decomposition

DiGress: Discrete Denoising diffusion for graph generation

Scalable Deep Generative Modeling for Sparse Graphs

Discrete-state Continuous-time Diffusion for Graph Generation

Disentangled Spatiotemporal Graph Generative Models

Neural Graph Generator: Feature-Conditioned Graph Generation using Latent Diffusion Models

Gradient-Based Spectral Embeddings of Random Dot Product Graphs