Reproduce “A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ryien Hosseini, Filippo Simini, Venkatram Vishwanath, Henry Hoffmann

DG-Gen introduces a deep probabilistic framework for generating continuous-time dynamic graphs, directly modeling temporal interactions as a joint probability distribution. The framework generates synthetic graphs with high fidelity, preserving both feature distributions and topological characteristics, and achieves state-of-the-art performance in inductive link prediction.

Key insights

  • Modeling temporal edges as a joint probability distribution, factored into conditional components, captures complex dependencies between nodes and timing.
  • The autoregressive sampling process allows DG-Gen to generate synthetic dynamic graphs of arbitrary size and duration.
  • An inductive design enables the framework to generate graphs with entirely new nodes, overcoming a limitation of many existing models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.