Paul McNicholas was born in Cork, Ireland, and raised in Dublin, Ireland. He was educated at Trinity College Dublin, where he studied mathematics, high-performance computing, and statistics. He is currently a Professor and Canada Research Chair in Computational Statistics at McMaster University, Hamilton, ON. His research focuses on statistical approaches to clustering, with a focus on clustering using mixture models. To date, he has written two monographs and over 100 research articles. Recent work includes approaches for higher order data as well as approaches for dealing with outliers. He has received some recognition for his work, including the Steacie Prize for the Natural Sciences. He is currently Editor-in-Chief of Journal of Classification and a member of the College of the Royal Society of Canada.
A complete accident. I was studying mathematics as an undergraduate at Trinity College Dublin. I was nearing the end of my first year and I planned to switch from mathematics into human genetics. That would have meant repeating my first year but I was willing to do that. My college tutor, the late Donal O’Donovan (a wonderful person who helped me tremendously), was happy to support the move but felt strongly that I should sit the mathematics exams anyway. The exam setup was that all or almost all of the weight for each 24-week course went on one final exam, and the exam schedule happened to put statistics one week later than the other five exams. I did not really look at the statistics material until that fateful week – I was not as studious as I might have been! When I started to look through the statistics course material, I wrote to the professor, Eamonn Mullins, and asked if I could chat with him – this was unusual in some respects because office hours did not really exist. That chat with Eamonn, who I came to know and admire, together with the exam schedule led to me sticking with mathematics so that I could pursue statistics as a profession. The way Eamonn explained it to me, the ability to work with real data was central to being a statistician and so the desire to be a statistician that seized me twenty years ago had data science baked in, even though that term was not used.
There are a few. One is how to deal with outliers in clustering. A lot of good work has been done, including some I was or am involved in, but I think that whole area is a bit of a work in progress. I think it is a difficult problem, on both a philosophical level and in practice.
The annual meeting and the journal are really important for me. I joined TCS in 2008, a few months after I began my faulty career in Canada. For family reasons, it was not until 2013 that I was able to attend a TCS meeting but some of my graduate students attended earlier meetings. I had heard great things from these students before I got my own first impression of a TCS meeting during the social gathering on the first evening. I distinctly remember being greeted enthusiastically by the inimitable Stan Scolve as I walked into the bar in Milwaukee where the social gathering was being held. The meeting itself was highly enjoyable and the atmosphere was really friendly. All the talks were about classification, and the questions asked were clearly intended to be helpful, which was nice! I have not missed a meeting since 2013. I also really like the journal, right back to my time as a graduate student. It remains the only journal that exclusively publishes classification work and I think it is very important for the field that such a journal exists. A conference where all the talks are about classification and a journal where all the papers are about classification, what more could one ask?
The idea behind Journal of Classification was to have a journal devoted to classification techniques. The first issue was published in 1984 but the decision to establish the journal was made the year I was born (1981), which gives me pause for thought. Comparing articles being published in Journal of Classification today to articles from the 1980s, one can see how research in classification has driven the direction of the journal rather than vice versa. I think that is how it should be, and I hope that will be the future of the journal. I mean, to move with the field. There are also some important practical initiatives that I think are important for the future of the journal. Some of these are already happening or about to happen. For example, progress has already been made on reducing the time from submission to first decision and the time from acceptance to publication. Soon, we will have a clear position on data availability as well as a requirement for code availability. I think these are all important steps for Journal of Classification.
I’m always happy to share my thoughts on this but, first, I must admit that I had the great good fortune to have some very strong students early on. For example, the first two Ph.D. students I graduated – Jeff Andrews and Sanjeena Dang (née Subedi) – have gone on to have successful faculty careers. I expect lots of folks could have done just as good a job supervising students like that, and to have supervised them so early in my own faculty career really helped me. And it is not just Jeff and Sanjeena, I have had many more really strong students in the meantime – too many to list. My approach has always been to encourage students to read good, recent work. I might give a Ph.D. student an initial idea, but I think it is much better for the student if the bulk of the ideas in the thesis are their own. It is not easy to do that, I have to say. It is probably easier to be more prescriptive and active in helping students do the work, but I firmly believe that is not in the best interests of the student. I think another important aspect of supervising students is a willingness to provide good feedback which, amongst other things, means honest feedback. That is also not so easy.