Scalable Generative Modeling of Weighted Graphs

Area of Science:

• Computational Biology
• Network Science
• Machine Learning

Background:

• Weighted graphs are crucial in various scientific fields, yet deep generative models often struggle with their complex topology and edge weights.
• Existing models for weighted graphs have limitations in capturing joint distributions of topology and weights.
• Nonlocal dependencies between graph edges and weights pose significant challenges for generative modeling.

Purpose of the Study:

• To develop an advanced autoregressive deep generative model for abstract weighted graph data.
• To address the limitations of current models in handling joint distributions of graph topology and edge weights.
• To create a scalable and computationally efficient method for generating weighted graphs.

Main Methods:

• Developed BiGG-E, an extension of the BiGG model, employing an autoregressive approach.
• The model learns a joint distribution over weighted graphs, effectively exploiting sparsity.
• Achieved efficient generation of weighted graphs with n nodes and m edges in O((n+m)logn) time.

Main Results:

• BiGG-E successfully captures complex distributions over weighted graphs.
• Demonstrated superior performance compared to existing methods in simulation studies.
• Experiments on benchmark datasets confirmed the model's scalability and computational efficiency.

Conclusions:

• BiGG-E offers a significant advancement in deep generative modeling for weighted graphs.
• The model provides a scalable and efficient solution for generating complex weighted graph structures.
• BiGG-E has broad applicability in fields reliant on weighted graph analysis.