The Statistics Group
The first statistician at the
The statistics group at NTNU is now by far the largest supplier of Master-level students in Norway. Much of the current research is motivated by challenges from other academic disciplines and real-world businesses.
Areas of research
Besides population dynamics this research also covers topics as
- evolutionary biology
- population genetics
- conservation biology
- functional genomics
An important activity is statistical modelling and analysis of data from genomics, where multiple hypothesis testing is a central research topic. Ongoing research also includes exact hypothesis testing concerning parameters of discrete distributions in the presence of nuisance parameters.
The main research topics include
- Design of Experiments (DOE)
- reliability analysis
- extreme value statistics
In reliability, focus is modelling and statistical inference in connection with repairable and maintainable systems and calculation of system reliability of structural systems.
In extreme value statistics, focus is estimation of extreme responses of dynamic structures and extreme value prediction from sampled time series.
The research in Design of Experiments (DOE) is directed towards projection properties of non-regular two-level designs.
Stochastic spatial and spatio-temporal modelling
Focus is on stochastic modelling of spatial and spatio-temporal phenomena and inference of the associated model parameters. Based on indirect observations of the phenomena Bayesian inversion with prior models of the type mentioned above is performed. Various types of Gaussian random fields and Markov random fields are mostly used.
Moreover, simulation algorithms, approximations and decision analysis for complicated spatial and spatio-temporal models are being studied. The research is inspired by challenges in characterization of petroleum reservoirs.
Research is directed towards speeding up algorithms for handling complex statistical problems. Special focus is given to Gaussian Markov random fields and applications of the approach INLA which makes it possible to avoid MCMC for doing Bayesian inference for latent Gaussian models.
Topics studied are characteristic functions and choice of smoothing parameters in kernel density estimation and methods for Monte Carlo computation of conditional distributions given sufficient statistics.