Trondheim Symposium in Statistics 2020

The symposium will take place September 25 at Scandic Lerkendal conference center Klæbuveien 127, Trondheim.

Invited speakers:

• John Tyssedal (IMF, NTNU)
• Claudia Battistin (IMF, NTNU)
• Jørgen Ødegård ( AquaGen, Trondheim)
• Ingrid Heggland (Universitetsbiblioteket, NTNU)
• Emily Grace Simmonds (IMF and Department of Biology, NTNU)

Organizing committe: Geir-Arne Fuglstad and Stefanie Muff

09:30 - 10:30 John S. Tyssedal (Presentation)
Coffee break
11:00 - 12:00 Claudia Battistin
12:00 - 13:00 Emily G. Simmonds

13:00-14:30 Lunch

14:30 - 15:30 Jørgen Ødegård
Coffee break
16:00 - 17:00 Ingrid Heggland (Presentation)

Presentations are about 40-50 min, and then there is time for questions. There will also be a couple of minutes break between talks to get water and coffee.

Dinner at Scandic Lerkendal

IMF, NTNU

Some perspectives on industrial statistics with a view towards applications and research

Real problems are often the catalyst for innovative research and many important statistical ideas and methods, from the t-test to advanced quality tools, have their roots in industrial problems. This talk will give a brief overview of the history of industrial statistics and its current state, mixed with (highly subjective) examples of applications and research in the field. Some perspectives of industrial statistics in the future will be given.

IMF, NTNU

Inferring network interactions in the presence of unobserved neurons and plasticity

Network architecture plays a crucial role in determining the function of real and artificial neural networks. Two popular working assumptions when attempting to estimate the architecture from data are: 1) the entire population of neurons is observed, and 2) the connectivity is stationary. Both of these assumptions are wrong for most systems worth studying. In order to assess the goodness of assumption 1), using mean-field methods, we studied how the unobserved neurons affect inferences of the connectivity between observed ones [1]. In the strongly connected regime, our results call for models accounting for the unobserved neurons — for which we developed an approximate expectation-maximization algorithm [2]. As for issue 2), we moved to a state-space model, where the non-stationary connectivity obeys a plasticity rule, which can be inferred via a particle filtering algorithm. This statistical framework provides an interesting perspective on the networks behind data and is particularly suited for brain networks, where plasticity is thought to form the basis of memory and learning.

[1] Dunn B, Battistin C. The appropriateness of ignorance in the inverse kinetic Ising model. Journal of Physics A: Mathematical and Theoretical. 2017 Feb 20;50(12):124002. [2] Battistin C, Hertz J, Tyrcha J, Roudi Y. Belief propagation and replicas for inference and learning in a kinetic Ising model with hidden spins. Journal of Statistical Mechanics: Theory and Experiment. 2015 May 19;2015(5):P05021.

Department of Biology and IMF, NTNU

Embracing uncertainty: a cross-disciplinary review of uncertainty consideration in statistical and mathematical models

It is well known that applied statistical and mathematical models will contain uncertainty. They are simplifications of reality and therefore will never perfectly capture the processes they seek to understand. This uncertainty does not undermine the utility of such models, providing it is correctly quantified and reported. But uncertainty can enter the modelling process from several different sources, which can accumulate as layers of uncertainty as model complexity increases.

Clear and transparent presentation of associated model uncertainties could even improve the usefulness of model results by indicating the plausible range of possibilities and allowing decision makers to include these in planning. As an increasing number of fields embrace a move to become more quantitative, it is becoming more important to undestand, quantify, and report the associated uncertainties.

Despite the ubiquity of uncertainty in statistical and mathematical models, reporting of uncertainty across fields is not standardised. In this project, we are developing a generalised framework that breaks down the key universal sources of modelling uncertainty. By creating a generalised framework that can be understood by researchers with any level of statistical training and applied across disciplines we can provide a standardised backbone from which to report uncertainty. With this we can begin to ask broad questions about how uncertainty is treated in different fields. We plan a systematic review of the proportion of published studies that report the different levels of uncertainty in each field, providing a snapshot of the state-of-the-art in uncertainty quantification and reporting. This project is in the early stages so I will present the motivation and background, the generalised framework, and the aims and scope of the rest of the project.

AquaGen, Trondheim

Principal components in genomic prediction

Genomic prediction is a recent methodology used in animal and plant breeding. Here, the genetic value of an individual is predicted using thousands of genetic markers all over the genome (typically 50’ or more), either by ridge regression on marker genotypes directly, or by animal models utilizing marker-based genomic relationships between the individuals. These methods have been shown to be more accurate than classical ancestry-based selection methods. However, markers closely located within the same chromosomal region tend to be inherited together and their genotypes are therefore often closely correlated. Predictive models using principal component analysis is well suited for analysis of this type of data and be used for dimensionality reduction, making genomic prediction models more computationally efficient. Examples of principal components analysis in genomic prediction models will be shown and discussed.

University library, NTNU

Open Science and Open Data – what and why?

This talk will start by introducing Open Science and some of the different aspects and initiatives included in this umbrella term. I will argue that Open Science in many ways is just science done right: responsible, reproducible and accessible. This move towards more transparent and reproducible research requires, to varying degrees, a change in culture and practice. There are barriers and challenges on the road towards openness, but also advantages and opportunities, which will be discussed. I will also go more in depth on the topic of Open Data and FAIR Data, which for many researchers is still a novel topic, but with great possibility.