Abstract:
|
Inference for models with intractable normalizing functions, such as some point processes and network models, poses serious computational challenges. We will discuss the relative merits and disadvantages of auxiliary variable Markov chain Monte Carlo (MCMC) and other approximate approaches that solve this problem. Each of these methods suffers from computational issues that makes it impractical in many settings. We propose novel algorithms that provide computational gains over existing methods, discuss some theoretical issues, and illustrate the practical application of our methods to real data examples.
|