Restricted Search Space Graph MCMC

A summary of my work on restricted search space MCMC for graph inference, using birth-death processes.

My primary research focuses on developing scalable MCMC methods for graph inference. This page summarizes the key contributions of my work, Restricted Search Space Graph MCMC via Birth-Death Processes, joint with Kieran R. Campbell and Radu V. Craiu.

Key contributions:

  • Sharp lower and upper bounds on the restricted search space error introduced by hybrid order MCMC methods
  • A novel trans-dimensional sampler that places a prior directly on the set of search spaces
  • A closed-form expression for this search space prior, which lets researchers explicitly control the tradeoff between posterior fidelity and computational scaling
  • An efficient implementation of the sampler in R, available at morrisgreenberg/RestrictedSearchMCMC
ROC AUC of recovering true graph edges with synthetic data generated from Gaussian Structural Equation Models, using a fixed sparsity budget.