Contributing Editor Xiao-Li Meng (who is also the new President-Elect) writes about two of the IMS special invited lectures he was inspired by at the Joint Statistical Meetings:
JSM has been a magnet for me since 1989—only once I had to depart on Tuesday due to a scheduling conflict. But JSM 2017 was the first time that I attended all three Wald lectures. The topic, “What’s happening in Selective Inference?” was undoubtedly a magnet in and of itself. However, the lecturer, Emmanuel Candès, made far more effort to keep the attendance at Wald III, despite it being at 8:30 am on Thursday, at a rate that is of the same asymptotic order as that of Wald I–II (but I was too absorbed to pin down the constant coefficients).
Emmanuel’s lectures truly realized the aim of the Wald Lectures, i.e., “to develop material in some detail and make it accessible to non-specialists.” His general introduction was the most comprehensive one I had ever seen at Wald Lectures and the like (and in this case, my non-ignorable missing-lecture mechanism only strengthens this evidence), exceeding 1/3 of his Wald I. He started with the headline news “Spaghetti sauce and pizza fight cancer” (was I ever happy to see the gluten-free option with the sauce only!), and reproduced much of the (media) cry on the irreproducibility crisis. Unlike some other discussions, Emmanuel’s message carried no tone of blame but only that of progress. He emphasized the positive response from the scientific community from a broad societal level to deep foundational research; and how the issue of replicability, a more profound one than reproducibility, had led to a new scientific paradigm. This then flowed seamlessly into a presentation on knockoffs—yes, that’s a technical term, just as is bootstrap—and its successful application in a genetic study, showing how it led to more and more replicable discoveries as confirmed by other studies. The audience was then hooked.
Once the audience was enticed to find out how the knockoffs were created in Wald II, Wald III then helped them navigate the rapidly evolving landscape of selective inference. (Emmanuel’s slides can be found at http://statweb.stanford.edu/~candes/Talks.html; incidentally, many recent talks and papers on selective inference are at http://www.math.wustl.edu/~kuffner/events.html). This survey of others’ work also frames the context of one’s own contributions, a telling characteristic of great scholarship. Emmanuel’s lectures fully showcase the multivalency of a first-rate scholar: caring for mentorship—there were frequent emphases on the contributions of his students and postdocs—and caring about education: Wald III ended with a passionate affirmation that “Education (undergraduate and graduate) will play a crucial role in communicating ideas and methods.” His special dedications, to a pioneer in the field (Yoav Benjamini, Wald I), to a loved one (Chiara Sabatti, Wald II), and to an inspiring colleague (Maryam Mirzakhani, Wald III), connected the audience on a warm and personal level, and reminded us once more to put our accomplishments in perspective: the pioneers’ shoulders, the loved ones’ arms, and the colleagues’ and students’ hands.
But the lingering perception of IMS lectures being too “mathematical” was still evident from an audience member’s response to Emmanuel’s humble line that he hoped that his lectures would not embarrass IMS: “What’s embarrassing to IMS is that I actually understood your lecture!” Hopefully, soon such a response would merely be a throwaway line. Indeed, I walked out of the Wald Lectures feeling very inspired and energized. This is what IMS lectures should be: Inspirational, Mathematical, and Statistical, all in one, and one for all. Being mathematical means being rigorous, explicit, logical, accountable and verifiable, attributes we should value more to ensure replicability; it is not at odds with being inspirational or statistical, as Emmanuel’s lectures vividly demonstrated.
Of course, no scientific claim can be made without at least one replication. Martin Wainwright’s Blackwell lecture was another powerful example of the ideal IMS style. Martin started with an insightful discussion of Blackwell’s wide-ranging contributions and then zoomed in on two trade-offs at the core of Data Science: computational versus statistical efficiency, and privacy versus the utility of data. He then presented deep and rigorous information-theoretic results in a way that both piqued the audiences’ interest and respected their intelligence. I particularly appreciated his emphasis on how a seemingly negative result about an “information barrier” actually led to a practical statistical method to overcome the barrier, because the precise mathematical result pinpointed the cause for the barrier.
As the President-elect of IMS (thanks for all your votes, especially considering you only needed a zero-truncated Bernoulli model to predict the outcome), I couldn’t feel more proud of IMS and optimistic about its future as a global leader in building the theoretical and methodological foundations for Data Science. Groundbreaking work has already been done by IMS members such as Emmanuel and Martin (and many more), and their IMS-style lectures can only attract and encourage many more young talents to join the force. So let’s venture together, IMS style!