Mathematical Sciences Seminar - Baysian social network analysis and applications
When:
—
Venue:
Birkbeck Main Building, Malet Street
No booking required
Recent advances in statistical computation has demonstrated the advantages and effectiveness of Bayesian approaches to social network data. Exponential random graph models (ERGMs) represent one of the most important families of statistical social network models.
The ERGM likelihood states that the probability of observing a given network graph is equal to the exponent of a vector of observed network statistics multiplied by an associated parameter vector divided by a normalising constant which is computationally intractable for all but trivially small networks.
Following the Bayesian paradigm, prior distribution is assigned to the parameter. Direct evaluation of the posterior distribution requires the calculation of the ERGM likelihood and the model evidence which are both intractable.
We present parameter estimation methods based on approximate Monte Carlo strategies for doubly intractable distributions which improve the efficiency of Bayesian methods for ERGMs and various applications from economics to neuroscience.
Contact name:
Department of Economics, Mathematics and Statistics