Timo Koski: "Understanding causal relations is needed everywhere"
Professor Timo Koski´s scientific interests are probability, genetics, causal inference and Bayesian networks. He is currently visiting FCAI with Professor Jukka Corander as host.
Can you introduce yourself and also share something surprising of yourself ?
Having been born, raised and educated in Finland I became recruited to Sweden, to Luleå University of Technology, to Linköping University and finally to KTH Royal Institute of Technology in Stockholm. These positions have been in the fields of applied mathematics or mathematical statistics.
Today I live in Lidingö, an island neighbouring Stockholm, except that now I stay in Helsinki at Töölö Towers, another nice place to be too. Something surprising: I am a member of Svenska Spårvägssällskapet which is a society for a bunch of nerds devoted to everything associated with tramways.
Your research focuses on Bayesian networks. How did you become interested in causality and causal inference?
I realized that very few researchers, say medicine, were happy about statisticians´ analyses producing correlations between a disease and some set of exposures instead of clarifying the causal relations between them.
Statistics alone cannot claim causality from data, but statistics and a causal model can. Bayesian networks are a language for talking about causal relationships and modelling them. They are graphical models expressing beliefs about connections between data by means of conditional probability distributions.
Why your research is important?
Causal thinking is a very basic form of human behaviour and activity. The ability to predict the effects of how we manipulate the external world is a basic requirement of mankind's survival. Causality seems to be inherently linked to counterfactual thinking like: what would have happened to my headache if I had not taken that pill?
The importance of this stand of research can be seen in that there are nowadays several research teams studying causal inference and its practical applications. The current discussion is frequently about the correct kind of scientific view of causality, no longer whether causal inference is a branch of science. I have experienced these comments when talking about Granger causality, a statistical notion based on prediction of effect by an impact.
How did you end up joining FCAI?
I have a longstanding collaboration with Professor Jukka Corander, who pointed out for me the possibility of applying for a fellowship grant at FCAI. I am very excited about this fellowship. I have been with FCAI in Helsinki since January 2020.
Last year I also noticed the importance of FCAI in a book named “Mitä tulisi tietää tekoälystä” written by Timo Siukonen and Pekka Neittaanmäki. This book is a fluently written well-informed survey with a certain special and welcome emphasis on AI in Finland and the Finnish contributions.
What do you expect from your time at FCAI?
I have been working with Jukka Corander on the Jensen-Shannon Divergence (JSD) and its applications to machine learning. JSD is a special measure of distance between two probability distributions. My feeling is that we have been making good progress.
Corander and his coworkers have applied adversarial sampling and JDS in a published study of vaccine-associated pneumococcal population dynamics. The data are isolates combined from available reference sequences in Massachusetts, Southampton, Nijmegen and the Maela refugee camp in Thailand. This is very important work, as it has bearing on antibiotic resistance.