|
Abstract:
|
When the calculation of the likelihood function is intractable, it is necessary to use approximations. This is a common problem in many application areas, for instance for population genetics models. Faced with this difficulty, two strategies can be considered. The first is to directly approximate the likelihood function and use conventional inference techniques based on this approximation. This is the approach taken in composite likelihood methods and in variational optimization schemes. Another possibility, in the field of Bayesian statistics, is to use intensive simulation techniques. Approximate Bayesian Computation (ABC) strategies belong to this class. The basic idea is to simulate new data values from the model and compare them to the observed data. This is an active research field that results in a rapprochement between Bayesian inference techniques and statistical learning methods. In this talk I will show how random forest techniques can be used to adapt ABC methodologies to large data sets.
|
