Nicolai Meinhausen is Professor of Statistics at ETH Zurich. His research interests include computational statistics, causality, high-dimensional data and machine learning.
He is currently Associate Editor for the Journal of Machine Learning Research and the Journal of the Royal Statistical Society, Series B.
During last year’s Joint Statistical Meetings in Seattle, Meinhausen presented the Medallion Lecture entitled: “Causal discovery with confidence using invariance principles”.
Alison Oliver talks to Professor Meinhausen about his career.
1. When and how did you first become aware of statistics as a discipline and what was it that inspired you to pursue a career in statistics?
I noticed during my time as a physics student how most sub-disciplines tend to become narrower the further you progress, but statistics seemed in contrast to involve many aspects from mathematics, computing and collaboration with other scientific fields. This seemed attractive (and still does today).
2. You are currently Professor in the Department of Statistics at ETH Zurich. Over the years, how has your teaching and research motivated and influenced each other?
Both teaching and consulting have shaped my research by reminding me that simplicity is more valuable than one would think in a pure research-environment.
3. Your lecture at JSM 2015 was the IMS Medallion Lecture II. Could you please tell us about your theme for the lecture and what points you brought across?
The title was “Causal discovery with confidence using invariance principles”, a topic I have been working on with collaborators for the last two years. Often we have data in practice that are neither carefully randomized (as in a clinical study) but are also not just purely observational. The data are instead collected in different environments such as data collected over different time-periods or under unknown interventions. The main point was that such data allow us to build valid confidence intervals for causal effects under certain general assumptions.
4. If there is one thing that you would like an audience member to take away from your lecture, what would it be?
Often we think of non-iid data as a nuisance. I wanted to show that on the contrary that so-called messy non-iid data can be very helpful for causal analysis.
5. Your research interests include computational statistics, causality, high-dimensional data and machine learning. What are you focussing on currently and what do you hope to achieve through your research?
I remain interested in a broader spectrum of ideas, mostly causality, machine learning and collaborations with people in physics at the moment.
6. What drew you to this specific field?
Causality is a good example for the point I was trying to make above: it involves philosophical aspects, mathematical challenges, computation and subject-specific knowledge and collaboration and can thus be a very varied and rewarding field of study.
7. David Harger at MIT commented to Forbes, ‘I’m sensitive to the fact that *all* machine learning algorithms tend to have errors, and am very interested in how we can keep humans in the loop in order to ameliorate the consequences those errors. This reflects a more general aspect of my research: I’m most interested in the computer as a tool that can help people be better at what they do, rather than as a tool that can replace people.’ Machine learning was amongst the hot topics at JSM 2015. What is the future of machine learning and what can we expect in terms of developments?
Dividing ML research into either “enabling people” or “replacing people” is interesting and something I have been conscious of for some time. Both streams of research are quite active at the moment. Of course, it is not always easy to predict which category a specific idea will turn out to fall in, but we can at least try to focus our energy on the first category, where much remains to be done in the analysis of large-scale data. Achieving a good tradeoff between statistical properties and computational speed is an interesting area of research. Having a faster method allows more iterations of a method and allows humans to play a more active part in the analysis process.
8. What would you say to encourage young people to follow a career in statistics?
It is really a unique field as it opens so many possibilities from more mathematical academic research to exciting work in science projects and many possibilities in start-up companies.
9. How do you manage to juggle Associate Editor roles on the boards of the Annals of Statistics Journal of Machine Learning Research, as well as your teaching and research commitments?
By also taking on an AE position for the Journal of the Royal Statistical Society…
10. Are there people or events that have been influential in your career?
I benefitted a lot from interaction with very interesting people in the field, including (but not limited to) Peter Buhlmann, Bin Yu, John Rice, Peter Bickel and Steffen Lauritzen. Working on many diverse applied projects in science and in consulting roles has always been challenging and a good balance to the more mathematical aspects of my work.
Copyright: Image appears courtesy of Professor Meinhausen