Intractable Likelihoods & Approximate Bayesian Computation (ABC) Workshop

intractable-workshop4x3.jpg

For statistical inference the methods based on the likelihood function, such as maximum likelihood estimation, REML and Bayesian inference, are usually those of choice.  However, such methods are not always straightforward to implement. For example, in complex problems often with large or massive data sets, it can sometimes be completely impossible to even evaluate the likelihood function. The computational statistics revolution of the 1990s provided powerful methodology for carrying out likelihood-based inference, including Markov chain Monte Carlo methods, the EM algorithm, many associated optimisation techniques for likelihoods, and Sequential Monte Carlo methods.

Although these methods have been and are highly successful in making likelihood-based inference accessible to a wide range of problems from almost every area of science and technology, they have their limitations.  For example, in high-dimensional problems and for massive data sets. Thus many challenging statistical inference problems of the 21st century cannot be addressed using standard likelihood-based methods. Examples which motivate the workshop come from network data, disease epidemiology, medical imaging, ecology, cell biology, climate extremes and airport design.

There have been various recent breakthroughs in computational and statistical approaches to intractable likelihood problems.  These include likelihood-free methods such as Approximate Bayesian Computation (ABC), composite and pseudo likelihoods, synthetic and indirect likelihoods, new simulation methods for hitherto intractable stochastic models, and adaptive Monte Carlo methods.

Presenters and tutors include:

• Professor Nial Friel, University College Dublin
• Dr Riccardo Rastelli, University College Dublin
• Professor Tony Pettitt, QUT and Chief Investigator at ACEMS

Outline program for workshop

Day 1 - Tuesday, 11 December

A gentle tutorial style day with R software and libraries.

You will be introduced to:

• examples of data and models with intractable likelihoods
• analytic likelihood approximations such as pseudo and composite
• indirect and synthetic likelihoods based on simulated data
• likelihood-free method Approximate Bayesian Computation
• computational algorithms MCMC, SMC
• an in depth example, analysis of network data and R library

Day 2 - Wednesday, 12 December

A  series of talks linking to Day 1 activities.  The participants include:

• Associate Professor Chris Drovandi - Modern developments
• Anthony Ebert - ABC for complex queuing models and airport design
• Professor Nial Friel - Modern developments
• Caitlin Gray - Simulation for network data
• Dr Markus Hainy - ABC model choice for spatial extreme data
• Dr Riccardo Rastelli - Network data models.
• Dr Matt Moores - Spatial data and Potts models
• Professor Scott Sisson - Modern developments
• Leah South - Bayesian synthetic likelihood: a parametric alternative to standard ABC
• Dr Brenda Vo - ABC for cell biology, implicit likelihood and agent based models