Deep RL Generates Graphs with Exact Assortativity Constraints
A new framework called the Deep Microcanonical Graph Generator (DMGG) has been developed by researchers, utilizing reinforcement learning to modify graphs through degree-preserving rewirings to achieve a specific level of assortativity, which reflects the correlation between the degrees of neighboring nodes. Unlike traditional exponential random graph models (ERGMs) that apply constraints in an expected manner, DMGG enforces strict constraints precisely. This research tackles a key issue regarding how the structure of a network influences its function by allowing for the effective sampling of microcanonical graph ensembles with controlled structural characteristics, extending beyond merely fixing the degree sequence. The paper can be found on arXiv with the identifier 2605.23285.
Key facts
- DMGG uses reinforcement learning for graph generation.
- It enforces exact assortativity constraints.
- Degree-preserving rewirings are employed.
- Contrasts with ERGMs that enforce constraints in expectation.
- Enables microcanonical graph ensemble sampling.
- Addresses structure-function relationship in networks.
- Published on arXiv with ID 2605.23285.
- Framework maximally alters graphs to reach target assortativity.
Entities
Institutions
- arXiv