Computational statistics
High-dimensional integration constitutes the core of many modern estimation problems. Effective computational algorithms make it possible to apply more realistic and interesting statistical models.
Main content
In order for statistical methods to be useful to practitioners, we need computer algorithms. The computational statistics group is developing numerical methods to solve problems of the type
argmax θ ∫f(x,θ) dx
where x is a high dimensional vector. This is the core of many estimation problems. Keywords for the research are 'Laplace approximation', 'importance sampling', 'optimisation', 'automatic differentiation', and the 'EIS algorithm'.
Applications
Our work is particularly focused on financial time series, non-linear Kalman filtering and generalised linear mixed models (GLMM).
Interested? Feel free to contact Hans Skaug.