Reproduce “Discovering Latent Network Structure in Point Process Data” 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
Scott W. Linderman, Ryan P. Adams
Researchers at Harvard University developed a probabilistic framework that infers hidden network structures from noisy point process data by integrating mutually-exciting Hawkes processes with flexible random graph models and Bayesian inference. This framework accurately discovered latent interaction networks in financial market trading activity and gang-related homicide data, outperforming alternative models in predictive tasks.
Key insights
- Decomposing the impulse response in Hawkes processes allows for separate modeling of network structure (presence/absence), interaction strength, and temporal dynamics.
- Data augmentation using latent parent variables simplifies the likelihood function, enabling efficient Gibbs sampling for interaction strengths and background rates.
- Integration of random matrix theory ensures that graph priors guide the model towards stable network structures, especially for increasing numbers of nodes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.