Large scale inference in the Infinite Relational Model: Gibbs sampling is not enough

The stochastic block-model and its non-parametric extension, the Infinite Relational Model (IRM), have become key tools for discovering group-structure in complex networks. Identifying these groups is a combinatorial inference problem which is usually solved by Gibbs sampling. However, whether Gibbs...

Full description

Saved in:
Bibliographic Details
Main Authors: Albers, Kristoffer Jon, Moth, Andreas Leon Aagard, Morup, Morten, Schmidt, Mikkel N.
Format: Conference Proceeding
Language:English
Subjects:
Bayes methods
Bayesian inference
Complex networks
Computational modeling
Gibbs sampling
Infinite Relational Model
large scale network modelling
Markov Chain Monte Carlo
Mathematical model
Mutual information
Stochastic processes
Citations: Items that cite this one
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The stochastic block-model and its non-parametric extension, the Infinite Relational Model (IRM), have become key tools for discovering group-structure in complex networks. Identifying these groups is a combinatorial inference problem which is usually solved by Gibbs sampling. However, whether Gibbs sampling suffices and can be scaled to the modeling of large scale real world complex networks has not been examined sufficiently. In this paper we evaluate the performance and mixing ability of Gibbs sampling in the Infinite Relational Model (IRM) by implementing a high performance Gibbs sampler. We find that Gibbs sampling can be computationally scaled to handle millions of nodes and billions of links. Investigating the behavior of the Gibbs sampler for different sizes of networks we find that the mixing ability decreases drastically with the network size, clearly indicating a need for better sampling strategies.
ISSN:1551-2541
2378-928X
DOI:10.1109/MLSP.2013.6661904