Reproduce “An Invertible Graph Diffusion Neural Network for Source Localization” 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
Junxiang Wang, Junji Jiang, Liang Zhao
The Invertible Validity-aware Graph Diffusion (IVGD) framework identifies original source nodes from observed diffusion patterns on graphs. It employs invertible neural networks and validity-aware layers to learn the inverse diffusion process end-to-end and ensure inferred sources adhere to valid graph structures, achieving over 20% average F1-Score improvement on diverse real-world datasets compared to baseline methods.
Key insights
- Invertible neural networks enable end-to-end learning of both forward diffusion and its inverse, providing a unified framework for source localization.
- Explicitly enforcing validity constraints on inferred sources is crucial for realistic and useful results, achieved through unrolled optimization layers.
- Linearization techniques within optimization layers can make complex constraint enforcement efficient and scalable for large graphs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.