A VARIATIONAL FRAMEWORK FOR GRAPH GENERATION WITH FINE-GRAINED TOPOLOGICAL CONTROL
Keywords: Controlled Graph Generation
Abstract: Controlled graph generation is the process of generating graphs that satisfy specific topological properties (or attributes). Fine-grained control over graph properties allows for customizing generated graphs to precise specifications, which is essential for understanding and modeling complex networks. Existing approaches can only satisfy a few topological properties such as number of nodes or edges in output graphs. This paper introduces CGRAPHGEN, a novel conditional variational autoencoder that, unlike existing approaches, uses graph adjacency matrix during training, along with the desired graph properties, for improved decoder tuning and precise graph generation, while relying only on attributes during inference. In addition, CGRAPHGEN implements an effective scheduling technique to integrate representations from both adjacency matrix and attribute distributions for precise control. Experiments on five real-world datasets show the efficacy of CGRAPHGEN compared to baselines, which we attribute to its use of adjacency matrix during training and effective integration of representations, which aligns graphs and their attributes in the latent space effectively and results in better control.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12497
Loading