— I think there are two factors that were the most important. On the one hand, ICEF gives you an opportunity to complete a joint program with University of London, and it is remarkable to combine both Russian math skills and rigorous British educational standards. On the other hand, I was amazed that ICEF hires faculty internationally. Having PhDs from top schools teach you economics is a valuable asset that other schools in Russia do not offer.
— You have become a leading scientific expert in machine learning. How do you think ICEF helped you to succeed as a scientist? Which ICEF professors most influenced your desire to become a scientist?
— Most importantly, ICEF teachers established my culture of research: they were the first to teach me how to ask important questions and how to write academic papers. ICEF provided me with plenty of research opportunities as a student. I enjoyed weekly ICEF research seminars, engaging in coursework, and writing my graduation thesis. ICEF really encourages you to do research, which does not happen even in the best US and UK schools. I think this is a great advantage for ICEF students.
I would like to thank Prof. Belyanin for supervising my course paper and Prof. Gelman for supervising my thesis. They were very passionate about working with me and spending their valuable time on our research. I think that without their support I would not have succeeded.
— You took a one year gap between graduating from ICEF and studying in grad school in America. Why did you make this decision, and what did you do during that year?
— I was not resting, but working on research and teaching. I was Appointed Lecturer for the Stochastic Calculus course at ICEF and Research Assistant at the LFE lab. My decision was to make a contribution to ICEF by teaching there and sharing my math knowledge. During that time, I wrote a research paper with Sergey Gelman on “Continuous Time Option Pricing with Scheduled Jumps in the Underlying Asset”. The substantial teaching experience that I gained at ICEF was crucial and valuable for my admission to grad school.
— The Courant Institute is the top-ranked applied math school in North America. How difficult was it for you to get in, and how did ICEF help you?
— Indeed, they have an extremely small acceptance rate and their admission decision is based on the math proficiency of applicants. It was very helpful that the math content of the ICEF program and its math courses are extremely strong, as the Courant Admissions administrators saw that all important math topics are covered by the ICEF curriculum. Also, the head of Courant Admissions, Prof. Bogomolov, personally knows about ICEF and recognises its high academic standards. Those two factors played out favorably in my admission.
— Tell us about your research. The SKPCA algorithm that you created in your thesis is cited a lot; what is it about?
— SKPCA stands for Supervised Kernel Principal Component Analysis. A kernel function (e.g. everyone knows the Gaussian kernel) is used to perform a nonlinear dimensionality reduction. The problem is to decide which kernel function to use. My algorithm fundamentally changes the game since it determines the appropriate kernel function automatically. It means that a human is no longer involved in the process, and the resulting precision is much higher. If you want to read more about the theoretical explanation, take a look at my published paper in JMLR.
— Why did you not proceed with a PhD degree at the Courant Institute?
— I think that the PhD is a great option for future scientists and sincerely encourage ICEF graduates to complete this degree. In my case, my qualifications were evaluated by Google Research as PhD level, and by the time I graduated my thesis “Generalization Bounds for Supervised Kernel Principal Component Analysis” was recognised by the research community. As a result, I was hired by Google Research immediately; thus, a PhD was not required for me. All of my colleagues at the Google Research lab have PhDs, and many of them have been full professors at universities.
— What have you been doing at the Google Research lab recently?
— While working there, I have developed a novel theory of coupled dimensionality reduction and algorithms that apply this theory in practice. It helped to improve Google products’ precision as well as reduce dimensionality reduction inconsistencies for a broad range of problems: from spam and malware detection to image recognition.
— Have you stayed connected to ICEF after you graduated?
— Apart from the fact that ICEF feels like a close family, I am excited to continue research collaborations with ICEF. In fact, I greatly appreciate my co-authorship with Sergey Gelman, as I think we have made a significant contribution. Moreover, we are working on additional contributions right now, 4 years after I graduated. We have recently been invited to present our paper at the NIPS Time Series Workshop in Montreal.
— If you go back in time, would you change your decision to study at ICEF?
— Definitely not. I am confident that ICEF has built solid academic foundation for me, and studying math and economics there was extremely valuable.
— What advice for current and future ICEF students would you give?
— Engage in research as early as possible, even when you are in your second year. Attend research seminars and try hard to complete your coursework with distinction. Also, focus on mathematics as early as possible, since all advanced courses in economics and other sciences are based on math; if you overlook some math areas in the beginning, it will be very difficult to catch up later. I even suggest that you take extra math courses in your first year and study with a tutor; the key areas to strengthen are analysis, linear algebra, and probability.
Oleg Seregin, for ICEF HSE