Working at the interface of statistics with applied and computational mathematics

Tagged: Machine Learning

My first conferences: A SAMBa story

  , , , , , ,

📥  Statistical Applied Mathematics

My first year at SAMBa has been quite an eventful one. I can definitely say that SAMBa has delivered on its promise to help me explore different areas of Mathematics. However, this post is not about maths at all (who are we kidding, it probably is), it rather concerns my experience of conference attendance. My very first one was a workshop on big models at the University of Warwick, which incidentally happens to be my undergraduate institution. The reason I went there was to see what the Institute of Data science was up to, and as expected, they were doing things that I was interested in. Statistics, applied mathematics, data science and machine learning.


Swisstech Convention Centre, Lausanne (courtesy of STCC)

My second conference was perhaps the most daunting one. In my quest to know more about uncertainty quantification (UQ), I took up a module with the local expert on it, Prof. Rob Scheichl. This led to the SIAM conference on uncertainty quantification in Lausanne. The venue was breathtaking, to say the least. It was in the Swisstech convention centre at the EPFL. I have never seen an auditorium with that much leg room during the plenary talks. To fund myself, I along with a fellow SAMBAlite, Gianluca, successfully applied for funding from MI-Net. Part of the deal of going there was to further inter university collaboration, and hence the day preceding the conference involved some informal chat with Fabio Nobile's research group. We exchanged ideas and research interests. The rest of the week involved us going to different sessions. They were quite daunting to be honest as most of the speakers were very, very good at what they did. This only increased my desire to actually start a research project and further my knowledge in a specialist area. The thing that I believe was most beneficial for me was to see the use of machine learning in UQ. This contributed to my decision to take up machine learning as my current research project as I could see that UQ would still be an option if I diverted away from it for a bit. I have to say though, my bilinguality helped a lot during the trip. I speak both French and English fluently, coming from a bilingual society (well, we're actually multilingual as most of us speak a third oriental language plus our local dialect, creole. Mauritius FTW right?).


Me, Gianluca and another SAMBa student Matt outside the conference in Lausanne

I went to my next conference after I had chosen my project (which is on automatic damage assessment in x-rays of Psoriatic Arthritis patients). It was a summer school on Gaussian Processes organised by the Machine Learning group at the University of Sheffield. It was the first year where they added the words "and uncertainty quantification" to it. I expected it to be full of computer scientists. To my surprise, the attendees were extremely diverse. There was an analytics team from formula 1; a guy working with Siemens (I think it was Siemens) on wind turbines; and wait for UQ lecturer from Warwick, Tim Sullivan (who is one of the most rigorous applied mathematicians I know) and my final year project supervisor, Mark Girolami. Uncertainty Quantification can mean many things! The summer school had practical sessions and talks from people who are very good at working with Gaussian Processes. One thing I got from the whole thing was that Gaussian Processes are used everywhere, from trying to optimise functions you cannot evaluate, to trying to fix the posteriors you get in Bayesian Inverse Problems. Is there a better alternative? Maybe. The search continues....


Antarctica as modelled using Markov Chain Monte Carlo, a key tool in Uncertainty Quantification (courtesy of the National Science Foundation)

The next conference I went to was very different from my two previous ones. It was actually one on Psoriatic Arthritis (PsA). No maths, just medical researchers and practitioners talking about, well, PsA. I did not get most of what was happening as they were talking about genes and things. They also mentioned "statistics" a lot of times. That was scary. I mean, really scary. When I asked them what kind of tests they were doing the answer was often "I don't know, I just press on a button in stata". It thus seemed to me that medical doctors in general need a better grasp of data science. Hence I tried to get some statistics training into their realm by suggesting we do a little session with consultants undergoing training. This was well received and it is hoped we can do it in the future during one of their training days.

You might be asking yourselves, how did you, a self proclaimed applied mathematician (please don't hate me, I know I did stats as a major) go to a medical workshop? Long story short: I was invited to a meeting in Bath on medical imaging, as I had now started working on this. There I met Prof. Neil McHugh, who works in pharmacology at the University of Bath. I mentioned my desire to know more about the PsA and he suggested I go there.

My English teacher always told me to put a concluding comment in my essays, and this is kind of an essay. The food at non-maths meetings, like the medical imaging one and the PsA one is way better. I had three course warm meals in both of them as opposed to the usual dried out sandwiches that repeat themselves n-times where n is the number of days the meetings stretch out for. Oh, and I'm writing this on a plane to Beijing where I transit before going to teach stats in Mongolia on SAMBa business. This one is for another time!