Trondheim Symposium in Statistics 2018

The symposium will take place Friday 28 – Saturday 29 September, 2018, at Bårdshaug Herregård.

Invited speakers: Sandra Hamel (UiT), Robert Jenssen (UiT), Juha Karvanen (University of Jyväskylä), Ingrid Van Keilegom (KU Leuven), Johannes Lederer (Ruhr-University Bochum).

Organizing committe: Bob O´Hara and Øyvind Bakke

14:00–15:15: Bus from NTNU in Trondheim to Bårdshaug. Check-in.
15:30–16:30: Johannes Lederer
16:30–17:30: Robert Jenssen
17:30–18:30: Guided tour of old building, aperitif
19–: Dinner and socializing

–9: Breakfast
9–10: Sandra Hamel
10–11: Juha Karvanen
11–12: Ingrid Van Keilegom
12–13: Lunch
13–: Bus from Bårdshaug to NTNU in Trondheim

Presentations are about 50 min, and then there is time for questions. There will also be a couple of minutes' break between talks to get water and coffee.

University of Tromsø

Quantifying eco-evolutionary processes using hierarchical, mixture and joint models

Populations are composed of varied individuals. The among-individual variation found within a population can affect eco-evolutionary processes population ecologists are trying to estimates. For instance, reproduction is expected to be costly, such that reproducing individuals should show reduced growth or lower survival. Nevertheless, some individuals do not seem to pay any reproductive cost and rather seems to always do better than others. With age, the individual composition of the population is unlikely to be a random selection of individuals, because frailty (i.e. weaker) individuals are likely to die earlier in life as a result of natural selection. Therefore, when trying to quantify life-history traits – e.g. reproduction, survival, or growth – and to estimate how traits change with age, we need to consider the influence of individual variance, autocorrelation, as well as covariance among traits. Through simulations and applied examples, I will illustrate how different statistical tools, namely hierarchical, mixture and joint models, can allow to account for these parameters and better quantify eco-evolutionary processes. I will also discuss the challenges and limitations evolutionary ecologists are facing when using these statistical methods.

University of Tromsø

Some advances in deep learning and kernel machines research

This talk will outline some main applied areas and research directions in the UiT Machine Learning Group. In the first part of the talk, research in deep learning (neural networks) for image analysis pixel classification is presented, specifically for handling imbalanced classes and for handling scenarios where data modalities may be missing completely or partially in the decision making process. In the second part of the talk, a new kernel (similarity measure) is presented for handling multivariate times series data with missing values. Finally, research at the interface of deep learning and kernel methods will be briefly presented.

University of Jyväskylä

Optimal observational design and multiple criteria decision making

In the first part of the presentation, we review typical design problems encountered in the planning of observational studies and propose a framework that allows us to use the same concepts and notation for different problems. In the framework, the design is defined as a probability measure in the space of observational processes that determine whether the value of a variable is observed for a specific unit at the given time. The optimal design is then defined, according to Bayesian decision theory, to be the one that maximizes the expected utility related to the design. Value of information can be described using the framework and the utility function can be generalized for multiple criteria decision making.

In the second part of the presentation, we propose using multivariate value of information in multiple criteria decision making. As a case study, we consider determining the appropriate conservation plan in forest management using data from south western Finland. There is an obvious trade-off between monetary and biodiversity criteria. To evaluate the conservation benefits, ecological inventories are valuable tools to accurately determine the potential benefits of the specific conservation area. However, the cost of an ecological inventory may exceed the value of information obtained. In our approach, the decision maker first makes an initial conservation decision and constructs an indifference curve that describes her preferences. Monte Carlo simulations are carried out to obtain multivariate value of information for the possible inventory schemes. The decision maker then chooses the ecological inventory scheme to be followed. The final conservation decision is made after the ecological inventory is conducted.

KU Leuven

Flexible parametric model for survival data subject to dependent censoring

When modeling survival data, it is common to assume that the survival time (T  ) is conditionally independent of the censoring time (C ) given a set of covariates. There are numerous situations in which this assumption is in doubt, and a number of correction procedures have been developed for different models. However, in most cases, some prior knowledge about the association between T  and C  is required. When neither prior knowledge nor auxiliary information is available, the application of many existing methods turns out to be limited. In this paper, we develop a flexible parametric model to estimate the association between T  and C, without any additional information. We show that the association between T and C  is identifiable. The performance of the proposed method is investigated both in an asymptotic way and through finite sample simulations. We also develop a diagnostic plot approach to assess the quality of the fitted model. Finally, the approach is illustrated on real data coming from a study on liver transplantations.

Ruhr-University Bochum

Tuning Parameter Calibration for Large and High-dimensional Data

Large and high-dimensional data has become a major source of knowledge in Economics, Biology, Astronomy, and many other fields. However, lasso and other standard methods for such data depend on tuning parameters that are difficult to calibrate. In this talk, we introduce novel approaches to this calibration and demonstrate their features in theory, computations, and applications.