SSubmitted to Statistical Science
A Conversation with JonWellner
Moulinath Banerjee ∗ and Richard J. Samworth † University of Michigan at Ann Arbor and University of Cambridge
Abstract.
Jon August Wellner was born in Portland, Oregon, in Au-gust 1945. He received his Bachelor’s degree from the University ofIdaho in 1968 and his PhD degree from the University of Washingtonin 1975. From 1975 until 1983 he was an Assistant Professor and Asso-ciate Professor at the University of Rochester. In 1983 he returned tothe University of Washington, and has remained at the UW as a facultymember since that time. Over the course of a long and distinguishedcareer, Jon has made seminal contributions to a variety of areas includ-ing empirical processes, semiparametric theory, and shape-constrainedinference, and has co-authored a number of extremely influential books.He has been honored as the Le Cam lecturer by both the IMS (2015)and the French Statistical Society (2017). He is a Fellow of the IMS,the ASA, and the AAAS, and an elected member of the InternationalStatistical Institute. He has served as co-Editor of Annals of Statistics(2001–2003) and Editor of Statistical Science (2010–2013), and Presi-dent of IMS (2016–2017). In 2010 he was made a Knight of the Orderof the Netherlands Lion. In his free time, Jon enjoys mountain climbingand backcountry skiing in the Cascades and British Columbia. ∗ Moulinath Banerjee is Professor of Statistics and Biostatistics at the University of Michigan,Ann Arbor, MI (e-mail: [email protected]). † Richard J. Samworth is Professor of Statistical Science and Director of the Statistical Lab-oratory at the University of Cambridge, UK (e-mail: [email protected]). imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 a r X i v : . [ s t a t . O T ] A ug M. BANERJEE AND R. J. SAMWORTH
Mouli:
Jon, I want to say that it’s a privilege to interview you. I would like tothank you for being a fantastic advisor and for your guidance, encourage-ment and inspiration throughout the years.
Richard:
To echo what Mouli says, it’s a great pleasure for me too!
1. CHILDHOOD
Mouli:
Can you tell us a bit about your family and background, and in partic-ular, your father Charles Wellner, to whose work you have a link on yourwebpage? In particular, how influential was he in fostering your love of na-ture and the outdoors?
Jon:
I was born in Portland, Oregon, and grew up in Missoula (Montana),Spokane (Washington), and Ogden (Utah). My father was a research forester;he started his career doing white-pine silviculture, and ended as a researchadministrator, organizing research centers for forestry in conjunction withUniversities in the intermountain west: Utah State (Logan), Montana State(Bozeman), and the U. of Idaho (Moscow). He had a huge influence on myinterests in outdoor activities. He was a skier during the 1920’s and 1930’sbefore organized ski areas were developed.
Richard:
Did your interests in quantitative fields start emerging during yourschool years? Any specific memories of school-life that you would like toshare with us?
Jon:
During high-school in Ogden (Utah), I did not focus on studies, but I didspend quite a bit of time skiing at the local ski area (Snow Basin). WhenI began undergraduate work at the U of Idaho, I was initially interested inpursuing a career in forestry, but started enjoying mathematics during myundergraduate work. So I switched majors after three years and ended upwith Bachelors degrees in Math and Physics.
Mouli:
Forestry’s loss was Statistics’ gain!
Mouli:
We remember a picture of you during your 65th birthday celebration aspart of a ‘scooter-gang’. Can you tell us a bit about that?
Jon:
Several of my friends during high school had scooters, and I also bought asmall Vespa to get to work at an ice cream factory in Ogden. The scootergang (Robert Johnstone, Steve Keller, and others) organized several longersummer trips during those years: once to Cedar Breaks National Monumentin Utah, and a longer trip to Yellowstone and Grand Teton National parks.I remember bucking a terrific head-wind in Wyoming during the return tripfrom Yellowstone.
Richard:
What a great photo!
2. UNDERGRADUATE YEARS
Richard:
You majored in Mathematics and Physics at University of Idaho. Howwas your academic experience there as an undergraduate? Did you considerother schools? imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Fig 1 . The intrepid and inimitable Vespa Gang. Which one is Jon?
Jon:
I started at the U of I in September 1963 after graduating from high schoolin Ogden in June 1963. My parents had both attended the U of I during the1930’s, so it was a natural place to go in terms of family history. The U ofI also had a scholarship incentive program that made it very affordable: ifyou earned a certain GPA (about 3.7 or above), then out of state studentswould only have to pay in-state (Idaho) tuition.
Mouli:
Since physics has its charms, especially to young quantitatively-orientedpeople, did you consider pursuing a career in that direction?
Jon:
I did, but I was more enchanted with mathematics at that point, andwas greatly enjoying the math courses, especially several courses taughtby Charles Christenson.
Mouli:
Do you remember which courses, specifically?
Jon:
The particular courses that have stuck in my mind as special were mathanalysis for two semesters based on the book by Tom Apostol, set theorybased on P. Halmos’s book
Naive Set Theory , and then a course on non-Euclidean geometry.
Richard:
How was the social life at U Idaho? Did you have lots of friends there?
Jon:
There was quite an active social life at the U of Idaho, but I was ratherfocused on academic activities during those years. I did manage to do a bit ofskiing and climbing while in Moscow, but that really increased substantiallyonce I started graduate school at the U of W in 1971.
Mouli:
We know that you served for a while in Vietnam. Was this immediatelyafter you finished your degree? Any reminiscences of Vietnam that you imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
M. BANERJEE AND R. J. SAMWORTH would care to share with us?
Jon:
Sure! The Vietnam war was going on while I was attending the U of I, so Ijoined the ROTC program and graduated with a commission as a Second Lt.in the Army with a two - year service obligation. I initially arranged a delayof service to begin graduate work, and I entered graduate studies at YaleUniversity in September 1968. At Yale I found it difficult to focus on studieswith the service obligation looming, so I left Yale in the spring of 1969 andbegan Army Service in June 1969. My first posting in the Army was to FortAugusta, Georgia, for a Signal Officers Basic Course. During that time theUS first landed on the moon. My nine month Army service with the SignalCorps in Nha Trang and Long Binh, Vietnam was fairly uneventful withthe exception of the occasional trip via helicopter to visit the signal sites onhill tops scattered over II corps (the second military region, stretching fromQui Nhon in the north to Phan Thiet in the south). I remember readingFeller volume I during that time. The US was trying to withdraw fromVietnam, so I ended up getting a three month “early out” from the Armyin March 1971. All things considered, I was very glad to leave the Armyand Vietnam in March 1971. I began graduate study at the Universityof Washington in an inter-disciplinary program in biomathematics duringspring quarter 1971 – starting with an undergraduate probability coursetaught by Albert Marshall.
3. GRADUATE SCHOOL AND ROCHESTER
Richard:
Tell us a little bit about the department at that time. What led you toempirical process theory and, in particular, to the topic of your PhD underthe supervision of Galen Shorack?
Jon:
The program in Biomath at the UW was very flexible. The strongest sub-group within that program was the new biostat group in the School ofPublic Health, but as students we had the possibility of courses in a vari-ety of departments, including the Math department. The probability andstatistics group within the Math department had quite a strong historyinvolving Bill Birnbaum, Ron Pyke, Bob Blumenthal, and Galen Shorack.Ron and Galen were both teaching statistics and probability during mystudent years. I got interested in the methods being developed by Galenand Ron in connection with asymptotic theory, and ended up doing a dis-sertation on some fairly technical problems connected with barrier crossingproblems for the empirical d.f. Along the way I caught the research bug:the problems formulated at the UW required a number of years of effort tosort out, but provided ample material for further research.
Mouli:
During your time at U Washington as a graduate student, did you get achance to interact with David Mason, Galen’s other stellar PhD student?Or was that later?
Jon:
Very much so! David and I were both in the advanced probability coursetaught by Galen during either 1972-1973 or 1973-1974. John Wierman wasalso a fellow student in that course. David did indeed also work with Galenfor his PhD, finishing a couple of years later in 1977 or so. Another studentin the program during those years was John Crowley, a student of Norm imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Breslow’s. The research by Norm and John on the large sample theory ofthe Kaplan–Meier estimator during Norm’s sabbatical year in Lyon wasquite intriguing, and was one of my motivations for learning more aboutweak convergence theory. Yet another student in the program who startedslightly later was Bruce Lindsay.
Mouli:
Who’s John Wierman?
Jon:
John Wierman was a fellow grad student who did a PhD with three differ-ent topics: a Berry-Esseen theorem for U-statistics, optimal stopping, andpercolation. John has had a distinguished career at Johns Hopkins Univer-sity, and was elected as a Fellow of the IMS in 1984. So that probabilityclass taught by Galen resulted in three Fellows of the IMS.
Richard:
In previous conversations, you have alluded to the early pursuit ofshape-restricted inference in the Pacific Northwest, which was eventuallyto become an important area of your own research. There were also peoplein Europe looking at this in the ’50s (Ulf Grenander, Constance van Eedenetc.) Can you tell us a little bit about these two parallel developments andalso how synergies were fostered between these different groups? Did yourearly exposure to shape-restricted inference happen during your graduatestudent years?
Jon:
Well, I was vaguely aware of work on shape restricted inference during mytime at the UW, but it certainly did not develop into a research interest un-til considerably later. On the other hand, the contacts between the PacificNW and the Netherlands was a major component of my awareness of inter-national connections and activity. Several statisticians at the University ofOregon (Fred Andrews, Don Truax, and (perhaps) Ted Matthes?) had spentsabbatical leaves at the Mathematisch Centrum in Amsterdam during thelate 1950’s and early 1960’s, while Willem van Zwet from the Netherlandshad spent a leave at the University of Oregon. Moreover, Galen, who didhis MS work in Math at the U of Oregon, had spent his first sabbaticalleave in Amsterdam. So I was well aware of the Dutch connection by thetime I finished my PhD at the UW in 1975. Part of the celebration of thatevent was a dinner at the Space Needle with Galen and Frits Ruymgaartwho was visiting the U of Oregon from the NL at that time. (Frits was, inan unofficial sense, Galen’s first PhD student.)
Mouli:
Ted Matthes incidentally was my former colleague, Michael Woodroofe’sadvisor at Oregon!
Mouli:
You joined the University of Rochester after your PhD from Washington.This is where you met one of your most important mentors, Jack Hall.Can you give us a brief account of your time at Rochester, and how Jackinfluenced your development as a researcher?
Jon:
Yes, I joined the U of Rochester in September 1975, mostly because of Jack.Just a brief story about getting recruited there. When the Boeing ResearchLabs closed in 1970, Al Marshall signed on for a one-year teaching positionat the UW; he temporarily replaced Ron Pyke who was on a sabbaticalleave during 1970-1971. Al then took a position at the U. of Rochester, andhe was there in the winter of 1975 when I interviewed for a job there. He imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
M. BANERJEE AND R. J. SAMWORTH was about to leave Rochester and return to the Pacific NW and a positionat the University of British Columbia. So I spent a few years following AlMarshall back and forth across the country. Joop Kemperman, a probabilistat the U of Rochester, was another important reason for going there. Jackwas very positive and open about research. I learned a lot from him aboutcontiguity theory, and we collaborated on several papers, including twopapers on mean residual life and another paper on confidence bands for theKaplan–Meier estimator.
Richard:
Tell us a bit about how the Begun, Hall, Huang and Wellner (BHHW)paper on semiparametric efficiency came about. Apart from Jack, who else,if anyone, influenced your work/interests in semiparametric theory at thattime?
Jon:
I was trying to understand contiguity theory and how it worked during thefirst few years of my career. Jack Hall and Bob Loynes had proved a resultabout the uniform integrability of likelihood ratios, and that provided astarting point for reading more of the work of H´ajek and Le Cam. (I havecalled their result “Le Cam’s fourth lemma”, but it really came from JackHall and Bob Loynes.) There was also the issue of asymptotic efficiencyof Cox’s “partial likelihood” which was addressed as a special case in aninteresting JASA paper by Efron in the late 1970’s. The challenge was todevelop a general approach which would handle both the Cox proportionalhazards model and the models stemming from Charles Stein’s work in the1950’s in which “adaptive estimation” was possible. In the late 70’s andearly 1980’s most of the focus was on those problems where “adaptivity”occurred. I found that several of the papers by Rudy Beran provided areadable entry point for some of the theory current in the mid-to-late 1970’s.I had some basic insights while preparing for a talk at Columbia in the springof 1980 that jump-started my own approach, and I pursued this during myinitial research work at the University of Munich while on sabbatical leavefrom Rochester during 1980 - 1981. Although I gave an initial talk aboutthat work at the Dutch meeting of Statisticians at Lunteren in November1980, it took 2 more years to get the paper written and on the way topublication.A side story about learning about people and their contributions: in Juneof 1980, before starting a German language course at the Goethe Instituteoutside Munich, I traveled to my first meeting in the eastern block, inBudapest, Hungary. During that meeting Willem van Zwet was going tohave dinner one evening with a young Russian fellow by the name of BorisLevit, and he invited me to join them. We had a very pleasant dinner in acastle above the river in Budapest, and it was clear that Willem was tryingto figure out how to get Boris out of Moscow and into a position in theNetherlands or somewhere else in the west. At that time I did not have aclue about the scientific background of Boris or his accomplishments. Quiteby chance while browsing in the math library in Munich later that Fall I ranacross a paper by Boris, and began to make the connections: it turned outthat he was doing (together with Yu A. Koshevnik and others in Russia) thesame kind of thing that I was trying to do in connection with the BHHW imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER paper, but from a different (nonparametric) perspective. Mouli:
The BHHW (1983) paper had a strong impact on the subsequent devel-opment of semiparametric theory in the 1980’s and 1990’s. How would youassess the legacy of that paper?
Jon:
The (1983) BHHW paper certainly got some of the simpler, broad brush,parts of the theory right, but there were many subtleties that were glossedover or even given a somewhat mis-leading treatment. My only excuse isthat I was struggling to learn enough math to formulate the problems cor-rectly. In another respect, the paper might have had a bigger impact ifwe had gotten the title right by somehow including the word “semipara-metric”. Many of the subtleties became more apparent during the work onBKRW (1993) over the next 10+ years with Peter Bickel, Chris Klaassen,and Ya’acov Ritov. In any case the main thrust of the BHHW paper wason target, and did have the effect of moving the focus away from the specialcases involving “adaptive estimation”.
Richard:
You spent your first sabbatical at the University of Munich under theauspices of a Humboldt Foundation Grant in 1980-81. Can you tell us a bitabout that experience? Did you, in particular, get to interact there withsome people whose academic ideas were influential in the future? Did youmeet Peter Gaenssler, whom you list as one of your mentors, at that time?
Jon:
Yes, the choice of Munich resulted from a correction to one of my earlypapers pointed out by Peter Gaenssler and Winfried Stute. Peter nominatedme for the Humboldt Fellowship, and the Humboldt funding resulted inthe year in Munich. The group there had a seminar going on martingaletheory, and that was interesting to me because of the work on the Kaplan–Meier estimator via martingale theory by Richard Gill and the Copenhagenschool. During the year in Munich I made visits to the Netherlands at leasttwice: during the Fall of 1980 Richard was away in Copenhagen workingwith Nils Keiding and Per Kragh Andersen, but I did succeed in trackinghim down in Amsterdam during the Spring of 1981. I should note thatPeter Gaenssler made very effective use of the Humboldt Fellowships overthe years, supporting not only myself in this way, but also David Pollard(before me) in Bochum before Peter moved to Munich, and then DavidMason a few years later (also in Munich).
Mouli:
Any other specific memories of Rochester? Michael Akritas once told methat he, you and Jack used to go skiing quite a bit at Rochester!
Jon:
Yes! The Wednesday night ski trips to Bristol Mountain south of Rochesterwere a regular event! Bristol had a thousand feet of vertical drop for skiing,and it had plentiful artificial snow during the cold Rochester winters, so Imanaged to do a fair amount of downhill skiing.
Mouli:
I thought Rochester gets a lot of natural snow as it is...they still neededartificial snow?
Jon:
But at a ski area the snow gets pushed around and often forms bumps ormoguls. Since it was often cold enough to make snow, the local ski areatook advantage of every opportunity to make more. imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
M. BANERJEE AND R. J. SAMWORTH
Fig 2 . A butterfly seeks the nectar of mathematical statistics.
4. 1983 ONWARDS: UNIVERSITY OF WASHINGTON
Richard:
I suppose one could say 1983 defines the beginning of a new phase ofyour academic career as you returned to the University of Washington andhave stayed there ever since. What prompted you to move back to youralma mater? Did the appeal of the Northwest play a big role in this?
Jon:
The appeal of the Pacific Northwest played a huge role. Proximity to familywas another major motivating force. By this time my father and youngerbrother had moved to Moscow, Idaho, for my father’s retirement and his“second career” on natural areas. The creation of the Statistics Departmentat the U of Washington in 1979 also played a huge role. During the time atRochester I visited the UW during the summers of 1977, 1979, and 1982.Those visits kept the UW connections going during the Rochester years.
Mouli:
Your first book on Empirical Processes with Galen Shorack was com-pleted in 1986, a relatively short while after you moved back to Washington.When did the two of you decide to go ahead with this project? How did itdevelop?
Jon:
Galen invited me to join him in writing the book toward the end of mysummer visit to the UW in 1977. Our work on the book really got goingin 1978-1979, but then it slowed down when I was on sabbatical leave in1980-1981. In any case, the whole project lasted for nearly 9 years, 1977 -1986. Our editor at Wiley, Bea Shube, was very supportive and kept urgingus on.
Richard:
Given the length of the book, that’s still a pretty decent rate of writing!
Richard:
It is interesting to note that your four books were published withina span of 10 years: your book with Galen came out in ’86, the one withAad van der Vaart in ’96, and your other two books in ’92 and ’93. Whatmotivated this very active book-writing phase of your career and how do imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER you see this in hindsight? Jon:
Well, that is difficult to explain. In part it resulted from a desire on mypart to understand the current state of limit theory in statistics and someof the unifying tools. As the process of doing research developed, it wasfairly clear that there were big gaps in the systematic coverage of variousareas, including empirical process theory. And it was also clear that generalempirical process theory – which was developing rapidly during the mid-1980s (thanks to the pioneering work of Dick Dudley, David Pollard, RonPyke, Evarist Gin´e, Joel Zinn, Mike Marcus, Jørgen Hoffmann-Jørgensenand others) would be extremely useful for all sorts of statistical questions.Fortunately, I managed to find excellent co-authors who were also interestedin some of these developments.
Mouli:
Tell us a bit about your famous book with Bickel, Ritov and Klaassenon semiparametric theory. How did the synergies that led to this book takeshape? You’ve of course written papers with all three of them, but how didthat book develop?
Jon:
The BKRW book resulted from an invitation that Peter Bickel receivedfrom Bob Serfling at Johns Hopkins University to give a series of lectureson estimation theory in the summer of 1983. Peter initially invited ChrisKlaassen and myself to join him in writing up lecture notes on semipara-metric theory based on his course at Hopkins. Shortly thereafter Ya’acovRitov began a series of visits to Peter, and they began solving a set ofbasic problems connected with the theory, so Ya’acov joined the projectas well. During my second sabbatical at Leiden in 1987 - 1988, Aad vander Vaart was just finishing his PhD work with Klaassen and van Zwet onsemiparametric theory, so Aad and I had many fruitful discussions aboutthe theory during the Fall of 1987, and some of the examples eventuallyfound their way into Aad’s very nice 1991 Annals paper. A number of pa-pers were generated during that time by just trying to figure out how thetheory interacted with several basic examples.
Richard:
Before we get to your next two books, let’s talk about the two scholarswith whom you wrote books, and with whom you have had very long terminteractions: Piet Groeneboom and Aad van der Vaart, and also about yourgeneral connections to the Dutch statistical school. Could you elaborate onhow your interactions with Piet and Aad, and in general, the Dutch schoolevolved?
Jon:
As I have mentioned above, my awareness of the Dutch school of statisticsand probability started developing during my time as a graduate studentin the early to mid-1970’s. I met Aad in Amsterdam at the 1985 ISI Meet-ing. Aad had started his PhD work in Leiden with Willem, and Willemintroduced me to Aad at that meeting. The meeting in Amsterdam was fol-lowed by a satellite meeting in Maastricht. I remember doing several walksaround Maastricht with Aad and Marie Huskova from Prague, during andafter that satellite meeting. imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH
Because Piet was in Seattle and Berkeley while I was in Rochester andMunich, my memory is that I did not meet him until 1987 or so, when Iwas on sabbatical leave in Leiden.Soon after going to Rochester in 1975 I attended the “Purdue Symposium”on statistics in the spring of 1975. Willem van Zwet was also attendingthe meeting, and we ended up sitting together for the conference dinner.It happened that Willem was very much aware of some of the results inmy PhD thesis, but he felt that I had tackled the wrong problem in somesense: I had proved a law of the iterated logarithm for linear combina-tions of order statistics, but that we did not yet have a good strong law oflarge numbers for such statistics. That suggestion took root with me andI went to Seattle for a few weeks and spent much of the time working outmy version of such a theorem. The result was published in the
Annals ofStatistics in 1977; Willem’s much more elegant and general paper on thesame topic was published a year later in the
Annals of Probability . In anycase, my connection with Willem has been a very important component ofmy relationship with the whole group of statisticians and probabilists inthe Netherlands, including Frits Ruymgaart, Chris Klaassen, Richard Gill,Piet Groeneboom, Geurt Jongbloed, and Aad van der Vaart. It has beena great honor to collaborate with all of them. In the end, the largest partof my collaborations have been with co-authors in the Netherlands. It hasbeen a great experience!
Mouli:
Was it primarily Piet who got you interested in the field of shape-constrained inference, the topic of this special issue, and an area that hasbeen a core theme of much of your research in the second half of yourcareer?
Jon:
Yes! I learned about Piet’s 1989 Rollo Davidson Prize paper in the mid-1980’s long before it was finally published in (1989). This paper remains atour-de-force benchmark in terms in the whole area. Piet’s paper provides acomplete and detailed description of Chernoff’s limiting distribution of thelocation of the maximum of two-sided Brownian motion minus a parabola.It illustrates the great value of focussing on a concrete problem. Piet hasreturned to this theme in the last few years, giving new proofs of the resultsin his ’89 paper in collaboration with Steve Lalley and Nico Temme. Thearea of shape constraints is building up a store-house of further problemsin this direction which await either the further research of Piet himself orthe interest of future research workers.
Richard:
Tell us about your book on Information Bounds and NonparametricMaximum Likelihood Estimation that was published by Birkhauser in 1992.What prompted the writing of that book? The first part of the book dealswith semiparametric theory, which is also covered by BKRW, the secondpart is largely on interval-censoring models (current status data and Case2 interval censoring). Would it be correct to suppose that your evolvinginterests in interval censoring at that time had something to do with thisbook? You subsequently wrote a number of interesting papers on the currentstatus model in particular.
Jon:
Piet was invited to give a series of lectures at G¨unzburg in southern Ger- imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Fig 3 . Left: Jon, Fadoua Balabdaoui, and Piet pose amid verdant surroundings in Seattle 2004;right: Jon wears a smile of infinite contentment, and Lutz strikes a smart pose for the camera(taken during a dinner at Noordwijk during Piet’s 65th Birthday Meeting in 2006). many in 1991 or 1992. Piet had been working on the theory of estimatinga distribution function with interval censored data during 1987, and wediscussed that work at Lunteren in November 1987. I spent time trying towork this model into the book with Bickel, Klaassen, and Ritov during thespring of 1988 and I had a number of exchanges with Piet about all thatin 1988 - 1989. So Piet invited me to join with him in writing up somenotes from his lectures. I ended up learning a lot about Piet’s methods andapproaches during that time, and that was my start on serious involvementwith shape-constraints. I was fascinated with the fact that the same limitingdistribution (non-standard; Chernoff’s distribution) was arising in at leasttwo quite different nonparametric monotone function estimation problems.It is now well-known that it arises in a large class of such problems, but forme in the early 90’s this was new and interesting.
Mouli:
It seems the Chernoff limit result for Case 1 censoring appears for thefirst time in this book? It was never published in a journal, was it?
Jon:
That is correct as far as I know. Piet had written a technical report at theUniversity of Amsterdam in 1987 that contained the result.
Mouli:
Let’s talk now about your most highly cited book (with Aad), whichas of going to press, has garnered 6000 citations. This book, by and large,introduced the general statistical audience to the tools of modern empiricalprocess theory, going beyond the more traditional empirical process the-ory covered in your earlier book, and clearly fulfilled a dire need. It wouldprobably not be amiss to say that this book is a standard toolkit of a largemajority of mathematical statisticians and theoretically-inclined method-ologists today. How did this book come about? And having witnessed itsgrand success, how do you feel with hindsight?
Jon:
The success of the book with Aad has been very gratifying. I learned anenormous amount from working on the book and from working with Aad.We started by trying to write down some of the basic theory needed to imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH
Fig 4 . Jon manages to distract Piet from his beer at St. Flour, 1994
Fig 5 . Aad and Jon relax at the Hortus Botanicus, Leiden 2015 develop several examples. One particular motivating problem was to justifythe approach that I had suggested in my 1989 discussion of an importantpaper by Richard Gill, “Non- and semiparametric maximum likelihood esti-mators and the von Mises method (Part I)”. Scand. J. Statist. 16 (1989), 97- 128. The issue was justification of Hadamard (or compact) differentiabilityin a general setting. To do that Aad and I needed a generalization of Her-man Rubin’s generalized Mann - Wald (or continuous mapping) theoreminvolving a sequence of functions { g n } rather than just one fixed continuousfunction g , to the Hoffmann-Jørgensen weak-convergence framework. This imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 NTERVIEW WITH JON A. WELLNER is just one example of problems in which we needed extensions of the clas-sical weak convergence theory. And in the end the simplest approach wasto write a book where all the needed theory could be collected in one place.[We also wrote a paper on our early efforts to understand the H-J theory,“Prohorov and continuous mapping theorems in the Hoffmann-Jørgensenweak convergence theory, with applications to convolutions and asymptoticminimax theorems”, which received rather negative reviews from the jour-nal where we submitted it at the time. It seems that most of the reviewersof that paper felt that the heavy theory was simply over-kill. That mighthave been true, but I would argue that it was very useful theory in thelonger term. For example, it has recently been useful in the study of multi-variate distributions via optimal transport theory; see e.g. Chernozhukov,Galichon, Hallin, and Henry (2017).]We had the good fortune of producing the book at an opportune time –when the theory had developed to a point where quite a bit was known, butthere also remained quite a few open problems. When we wrote the book thetheory of concentration inequalities was still under very active development.There are several other instances of areas that were still developing quitequickly, and which did not make it into the 1996 book. Mouli:
The recent Boucheron, Lugosi, Massart book on Concentration Inequal-ities probably fulfills the same sort of need that your book on EmpiricalProcesses did at that time!
Jon:
Yes, I agree. Of course the book by Boucheron, Lugosi, and Massart focuseson inequalities per se, while the 1996 book with Aad covers other groundas well. The success of the Boucheron, Lugosi, and Massart book is ampletestimony to the importance of inequalities!
Richard
Is a revision of
Weak Convergence and Empirical Processes forthcomingsometime in the near future?
Jon:
Yes, at least we hope so! Aad and I sent a revision to Springer in Octo-ber last year. Unfortunately we have not received a definite response fromSpringer yet. I continue to hope that Springer will publish a revision withinthe next year. [The revision does include quite a few improvements andadditions, so it will be well worth buying a copy when it does come out!]
Richard:
That’s excellent news!
Mouli:
Count me in! I already have two copies, one for the home, the other forthe office... and maybe the new edition I’ll put in my travel bag!! Ha, ha!
Mouli:
You had an active collaboration for many years with Norm Breslow atWashington, who was also a close friend. Presumably, this was also influen-tial in triggering your interests in models with Biostatistical applications.Would you tell us a bit about your work with Norm?
Jon:
As I mentioned above, my first interactions with Norm came indirectly viaJohn Crowley and their joint work on the large sample theory of the Kaplan-Meier estimator. When I got back to Seattle and the UW, I graduallygot connected with a group of ski-mountaineering folks through Norm andothers in the Math Department. And then Norm organized trips to Nepal in1989 and 1996. In between Norm and I did quite a lot of climbing together - imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH - including successful climbs of Sloan (1987), Columbia (1992), Formidable(1995), Fernow (1997), and unsuccessful attempts to climb Jack Mountain(1993, just as his first grand-child was about to be born), Bonanza Peak, andDumbell. We also joined others in about 8 or 9 spring ski-mountaineeringtrips to the British Columbia Coast Range, starting in the early 1990’s andlasting until about 2013.Our collaborative scientific work only began in the mid 1990’s when Normgot me interested in two phase designs and the resulting statistical issuesarising in connection with these designs and semiparametric models. Be-tween 1995 and 2015 we wrote 5 or 6 papers on this topic, and my interestin this area has continued since Norm’s death in 2015.
Richard:
The theme for your 65th birthday celebrations in Seattle in 2010 in-volved the term ‘From Probability to Statistics and Back’. You have men-tioned to us on certain occasions how you have always enjoyed working atthe interface of probability and statistics, being able to drift from one tothe other and back. To what extent has this informed your research? Doyou feel that there is not that much of this happening anymore, now thatprobability and statistics have diverged somewhat, as disciplines?
Jon:
The freedom to go back and forth between parts of probability theory andstatistics has been very important for me. It seems that these connectionsare less important for many people working in only one or the other of theseareas, but I have enjoyed being able to spend time learning different bitsof probability theory and using them to help address statistical problems.This type of activity is still going on, but perhaps at a reduced level and insomewhat different directions than when I started my career.
Mouli:
You have had numerous students over the years working on the differentareas of your interest, more than 30 counting the ones who are still workingwith you. And much of your core work is contained in their dissertations.Tell us about your student-advising experience a bit, and how you havefound it rewarding.
Jon:
Supervising PhD students has been both rewarding and challenging. Everystudent is different, so the trick is to try to find the right match betweenthe student and the problem(s). Many students already know more or lesswhat they want to do and are very capable. I have been very fortunate inhaving supervised a number of very strong and creative students. And thenit is often the case that I end up learning more from them than they learnfrom me!
Richard:
I agree completely!
Richard:
How about post-doctoral supervision?
Jon:
I have had very little grant support available for postdocs over the years,and hence I have only supervised two post-doctoral students: Hanna Jankowski(from Toronto), and Adrien Saumard (from Rennes and Pascal Massart’sgroup in Paris). Both of those experiences were very positive and enjoyablefrom my point of view. Hanna and I studied estimation of convex hazardfunctions and estimation of discrete monotone distributions (resulting in imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Mouli:
Would you like to mention some of your own favorite papers? Maybe thetop five in your view?
Jon:
Sure! My favorite 5+ papers include: • my 1978 ZfW paper on “ratio limit theorems”. (I believe that JackKiefer was the AE for ZfW at the time and handled this paper.) • the BHHW 1983 Annals paper mentioned earlier; • the 1988 Annals paper with Richard Gill and Ya’acov Ritov on biasedsampling; • the 1993 Ann. Prob. paper with Jens Praestgaard on bootstrappingwith exchangeable weights; • the 2008 Annals paper with Leah Jager on goodness of fit tests basedon R´enyi divergencesSeveral of the papers on shape-constrained estimation should also be listedhere, but we can return to those later.
Richard:
Apart from Jack Hall and Peter Gaenssler, would you like to tell usabout some of your other major academic influences that have not beencovered in our conversation thus far? Maybe Evarist Gin´e, Richard Gill,Lutz D¨umbgen?
Jon:
Richard was very influential in getting me going on martingale theory inconnection with survival analysis. His work on the Cox model with PerKragh Andersen played a big role in getting solid large sample theory settledfor the Cox model. Richard’s 1989 paper on “Non- and semiparametricmaximum likelihood estimators and the von Mises method” convinced meof the importance and utility of Hadamard differentiation. (I have writtentwo or three papers with Richard over the years, and have greatly enjoyedthat collaboration.)(The late) Evarist Gin´e and Joel Zinn were two very active participants inthe “High - Dimensional Probability” group starting back in the 1970’s. Istarted attending the meetings of this group in the early 1990’s and orga-nized one of the meetings, HDP II, in Seattle in 1999.The papers by Gin´e and Zinn on general empirical process theory in the1980’s and early 1990’s were inspirational in terms of their scope and gen-erality. Evarist became a great friend after a visit to Storrs in 1996. Weonly wrote two papers together, but it was always a great pleasure to meetup with him at the HDP meetings or elsewhere and talk about empiricalprocess theory. He passed away far too soon.Lutz D¨umbgen and I have had a lot of fun working on several differentproblems, including a great collaboration with Sara van de Geer and MarkVeraar to better understand Nemirovski’s inequality, proving a version ofMarshall’s lemma for convex density estimation, proving a neat law of the imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH iterated logarithm for Grenander’s estimator, and ... figuring out new non-parametric confidence bands for distribution functions related to an intrigu-ing test statistic due to Berk and Jones. That last project is still “underrevision”, but it will be a very nice paper when it is finished! Lutz intro-duced both Vera and me to swimming down the Aare in Bern. I am prettysure that Vera would not do that again, but I might be up for another goat it – in the summer when the river has warmed up!
Mouli:
I remember helping out with the registration desk at the 1999 HDP IImeetings in Seattle, as a graduate student!
Fig 6 . A glimpse of (part of) an academic family: from left to right are Bin Nan (S) , NilanjanChatterjee (S), Bodhisattva Sen (Jon’s academic grandson), Mouli Banerjee (S), Galen Shorack(Jon’s advisor), Marloes Maathuis (S), Florentina Bunea (S), Roy Han (S), Takemi Saegusa(S). (‘S’ means Jon’s student)
5. SHAPE-CONSTRAINED-INFERENCE
Mouli:
Let’s talk a bit more about shape-restricted inference, since it has beena major theme of the later part of your career. What would you say mostdrew you to the area of shape-constrained inference? What is it about thearea that you particularly like?
Jon:
I have been attracted to shape-constrained inference by the nonstandardnature of the limit theory together with the large number of open prob-lems. The strong cross-connections with inequalities and convex analysis isanother attractive feature in my view. There are further cross-connectionswith probability theory via the non-standard limit theory, which we still un-derstand only partially. Piet Groeenboom’s 1989 ZfW paper illustrates thepossibilities in this direction. It is a very rich area with lots of opportunitiesfor both application and further theory.Another attraction has been the personal connection to particular peopleinvolved in developing the theory. For example, in 1976 or 1977, I attendeda regional meeting at Cornell on Stochastic Processes and Applications , ZfW=Zeitschrift f¨ur Wahrscheinlichkeitstheorie und Verwandte Gebiete, now known asPTRF = Probability Theory and Related Fields imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER with some sessions on statistics as well. I remember Jack Kiefer giving atalk about his 1976 ZfW paper with Jack Wolfowitz on what came to beknown more generally as “Kiefer–Wolfowitz theorems”.Let F n denote the usual empirical d.f. and let (cid:98) F n denote the maximumlikelihood estimator of a concave distribution function (i.e. the distributionfunction corresponding to the Grenander estimator of a monotone density).By building on a key lemma due to Al Marshall (appropriately enoughknown as Marshall’s lemma), Kiefer and Wolfowitz (1976) proved, with (cid:107) · (cid:107) denoting the supremum norm, that (cid:107) F n − (cid:98) F n (cid:107) = O (( n − log n ) / ) almostsurely under curvature hypotheses on the true d.f. F . When Lutz D¨umbgenand Kaspar Rufibach and I proved an analogue of Marshall’s lemma forconvex decreasing densities in 2007 (just in time for Piet’s 65th birthdayconference in Leiden), I knew that it should be possible to prove at least apartial analogue of the Kiefer–Wolfowitz theorem in this case for the leastsquares estimator. Fadoua Balabdaoui and I managed to accomplish thatin a paper that appeared in the same Festschrift volume for Piet. The storyis not over, though. These results remain incomplete and somewhat unsat-isfactory: we still lack a good Kiefer–Wolfowitz theorem for the MaximumLikelihood Estimator of a decreasing convex density, and we also still lacka Kiefer–Wolfowitz theorem for the log-concave MLE on the line, much lessfor the MLE of a log-concave density on R d . Richard:
Yes, recently Arlene Kim, Aditya Guntuboyina and I managed to provea version of Marshall’s inequality for univariate log-concave density estima-tion, but as you say we don’t have a Kiefer–Wolfowitz theorem.
Richard:
How many of your students and post-docs have worked on this area?
Jon:
To date 9 of my 29 past PhD students and both of my current PhD studentshave worked in this area for a total of 11 PhD’s. This represents a bit morethan one third of my past and present PhD students. And, of course, bothof my post-docs!
Mouli:
What do you think has spurred the surge in interest in shape-constrainedinference over the last decade? Do you think the close connections to convexoptimization, and more broadly, convex geometry have helped get a broaderaudience interested?
Jon:
Yes, the connections to convex optimization, convex geometry, and convex-ity based inequalities have helped to spur the increasing interest in shape-constrained approaches. I am still working toward a better understanding ofthe collection of inequalities connected with the Brunn–Minkowski theoryas outlined in the wonderful survey paper by R. J. Gardner (2002).
Richard:
What do you see as the next main challenges of the shape-constrainedcommunity?
Jon: imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH many fundamental questions about these estimators, not to mention infer-ence procedures beyond estimation.
Mouli:
Is the shape-constrained community having as much impact on applica-tions as it should?
Jon:
No, probably not yet. This is partially due to the difficulty of the theory andlack of readable expository and review material in the area. The recent bookby Piet and Geurt
Nonparametric Estimation under Shape Constraints: Es-timators, Algorithms and Asymptotics , might provide some push for chang-ing that. Development of faster computational methods might well play animportant part in increasing the number of applications, but the commu-nity also needs to do more work to provide inferential methods beyondestimation.
Richard:
Of the results in shape-constrained inference that you have established,which ones are your favorites?
Jon:
Among my favorites are: • the 2001a,b Annals papers with Piet and Geurt; • the 2001 Annals paper with Mouli; • the 2009 Annals paper with Kaspar and Fadoua on the pointwise limittheory for log-concave maximum likelihood estimators on R .I especially like the limit theorem for the log-concave mode estimator in thelatter paper!
6. THE BROADER PROFESSION AND HONORS
Mouli:
You have been involved in service to the broader profession at severallevels. In particular, you served as Co-Editor of the Annals of Statisticswith John Marden from 2001-2003. Would you tell us something aboutyour experience?
Jon:
Editing the Annals was a marvelous educational experience. It was morework than I had anticipated, but it provided a wonderful overview of thebreadth and depth of the current research interests of people all over theworld at that time. It was also quite a broadening experience in terms ofworking with a fairly large group of excellent people as Associate Editors.The other two journals I edited for the IMS,
Statistics Surveys and
Statis-tical Science , were also educational and broadening, but in very differentways than the
Annals experience: they are simply very different journals.
Statistics Surveys was just getting underway at that point, and in my viewis still underused and probably under-rated. The probability side of ourcommunity has been ahead of the statistics community in terms of making imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Fig 7 . Jon traces out complex shapes in the snow that will require formidable future analyses. good use of their version,
Probability Surveys , as well as being ahead interms of electronic venues for publication more generally.
Richard:
How, in your view, has the Annals evolved from that point till now?Any suggestions you have for the journal, going forwards?
Jon:
The Annals has grown and changed quite a bit since I was co-editor withJohn Marden. John and I were receiving about 300 papers per year and hadinitiated the possibility of electronic submission (which quickly became thenorm). We had about 25 Associate Editors and were still using a database system created by John Rice when he was a co-editor with BernardSilverman. My understanding is that, now, the number of submissions hasincreased to around 700 per year, and the number of Associate Editors isup to about 50. The whole submissions and review process for the Annalsand all the IMS journals is now handled through EJMS. Whereas the talkon the street in the early 2000’s was of “theory going away” and the Annalsof Statistics closing shop, exactly the opposite has occurred! The era of“big data” and “data science” have created challenging new problems andcreated the need for many further theoretical developments to make senseof all the new methods being developed.The IMS has a history of creating new journals when the need arises: the imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH first IMS journal, the
Annals of Mathematical Statistics , was the sole IMSresearch journal until 1972 or 1973 when we created two new journals, the
Annals of Statistics and the
Annals of Probability . Jack Hall was chair ofthe IMS Committee which recommended this split; Ingram Olkin was theinaugural editor of the
Annals of Statistics , and Ron Pyke was the inauguraleditor of the
Annals of Probability . The next split came in (1991) when theIMS created the
Annals of Applied Probability (with J. Michael Steele asinaugural editor). A further split came in (2007) when the IMS created the
Annals of Applied Statistics (with Bradley Efron as inaugural Editor-In-Chief, and three different area editors). Perhaps the time has come for yetanother new journal in the area of Data Science. In fact an IMS Committee(with Liza Levina as chair) is studying this possibility now.
Mouli:
You have also served as IMS President very recently. How was thatexperience for you?
Jon:
It has been enlightening, rewarding, and considerably more work than I hadanticipated. Since I am still somewhat “in the harness” as Past President(at least until the Annual Meeting this year in Vilnius), I won’t say anymore about that right now.
Mouli:
The field of statistics has clearly changed a lot since you started yourcareer. In particular, it now falls under the bigger umbrella of data sciencealong with certain streams of engineering and applied mathematics. In onesense, this is good as it enhances synergies and scope. On the other, thereis also the possibility of a loss of identity. Any thoughts on what an optimalcourse for the discipline would look like? More generally, what are yourperceptions of the discipline as you see it, today?
Jon:
As a discipline or field, statistics is still fairly young, and it has indeedchanged quite a lot during the span of my career. The primary driver ofthis has been the enormous changes in computing power which have oc-curred over that time span. Statistics clearly needs to keep working to notonly provide new methods for the many new applications arising in variousfields of science, but also ways of understanding the properties of the newmethods. This is likely to require quite a lot of new mathematics as well asnew statistics and new ways of organizing statistical theory to tackle thenew problems. As a discipline we need to be open to different ways thatindividuals and groups can contribute to research.
Richard:
One topic that we have had conversations on a number of times is‘reproducible research’, which is quite critical to keep the discipline on asolid honest footing. Do you think statisticians are meeting the bar whenit comes to this in general?
Jon:
No, probably not yet. “Reproducible” has a number of possible meanings inthe context of the field of statistics: all the way from documenting programsso that individuals can replicate their own computations a few years aftercompleting them, to the conduct of scientific investigations in a way thatleads to the same conclusions from different labs or groups. imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Mouli:
On July 30, 2010, on the occasion of your 65th birthday celebrations,you were made a ‘Knight of the Order of the Netherlands Lion’. Given yourlong term connections to and involvement with the Dutch School, this musthave been special to you. Perhaps you could tell us something about this?
Jon:
Yes, that has been very special: quite an unexpected honor. Willem vanZwet was the primary originator of this of course, but it certainly entailedsupport from many of my Dutch friends and collaborators, Piet, Aad, Geurt,Chris, Richard, and more. As far as I know, Willem has organized Knight-hoods for two statisticians in the US (Peter Bickel and myself), and two inEurope (Marie Huskova and Sara van de Geer). I confess I was caught com-pletely off-guard when this happened in 2010, but in retrospect I shouldhave realized from Willem’s hints that I should be wearing a tie for theconference dinner that something was up.
Mouli:
Any thoughts of retirement? You appear to be enjoying your research asmuch as at any other time that we have known you!
Jon:
I am still enjoying research work quite a lot, and my intention is to continuethat involvement for some time. But my current plan is to retire from myteaching position at the UW in 2020. I will take one more sabbatical leaveduring the Winter and Spring of 2019, then teach one more year (2019-2020)before retiring sometime between June and September 2020.
Fig 8 . Jon proudly wears the Emblem of the Order of the Netherlands Lion at UW, Seattle,2010
7. PERSONAL LIFE AND INTERESTS
Mouli:
You were part of the Mountain Rescue Team in Seattle, something thatdove-tailed nicely with your passion for mountaineering. I still rememberthe pager you used to carry around in the department. Tell us somethingabout your experiences as part of that team.
Jon:
I got involved in Seattle Mountain Rescue through Vic Ericson, the brotherof a friend from my time in the Army and Vietnam, Paul Ericson. Vic andI climbed together quite a bit during my graduate school days, and then imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH again at the Gunks in New York during the time I was in Rochester and hewas with ATT in New Jersey. Vic started working as a lobbyist for PNWBell when he returned to Seattle, and he got roped into SMR by anotherlobbyist and long-time SMR member, Bill Robinson.The missions with SMR varied enormously, from searches for missing chil-dren, to rescues of people with broken legs, and to straightforward bodyrecovery situations. It frequently involved a push to get to victims of anaccident as rapidly as possible. I found it to be a challenging activity inwhich one could sometimes make a real impact in terms of getting an in-jured person or party to safety. The most satisfying missions involved ac-tually getting someone who had been injured or lost out safely. Serving asa rescue member of SMR involved quite a bit training and practice time(learning the rigging systems and relearning first aid and communicationskills), but with a committed group of people who were often quite differentfrom my academic colleagues. I participated actively as a “rescue member”of the group from the mid-1980’s until the late 1990’s, and was involved inabout 50 missions over that time period. I also edited the newsletter, the“Bergtrage”, for SMR for about 10 years.
Richard:
You met your wife Vera through Mountain Rescue, isn’t that correct?And you both share an avid passion for mountaineering!
Jon:
Yes; Vera joined SMR just a little before I did in the mid 1980’s. Her con-nection was through the Climbing Committee for the Seattle Mountaineers.(The Climbing Committee is the group within the Mountaineers that orga-nizes the Mountaineers’ climbing courses.) We met on a mission to searchfor a missing skier at the White Pass ski area during March, 1986. Duringthe drive down to White Pass we had time to discuss the pros and consof the types of climbing we each enjoyed the most: she was into climbingelegant lines on solid rock, and had made several trips to Yosemite withfriends from the Mountaineers, while I was focussed more on peak bagging(which can involve inelegant lines with lots of unstable loose rock).
Richard:
You have been to Nepal several times. And some of these trips werewith Norm Breslow. Would you recount some of your experiences in Nepal?
Jon:
All of the Nepal trips were with Norm. The trips to Nepal in 1989, 1996,and 1999 were wonderful experiences. The first trip (1989) was with NormBreslow and David Thomas, and involved a “tea-house” trek around An-napurna over a period of three weeks, with one high pass, the ThorongLa, at an elevation of 17,669 feet. On that trip, Norm visited the site ofan enormous avalanche on the shoulder of Dhaulagiri which killed a Stan-ford classmate, Bill Ross, in 1969. (See “American Dhaulagiri Expedition -1969”. American Alpine Journal. American Alpine Club. 17 (1): 19. 1970.)David Thomas was involved in eradicating smallpox early in his career asan epidemiologist, a project which took him to India, Pakistan and othercountries in Asia and the Middle East.The second trip in 1996 was with Norm Breslow, another “Reedie” PeterRenz (both Norm and Peter did their undergraduate studies at Reed Collegein Portland, Oregon), and Rob Schaller, a friend of Peter’s who had been imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Fig 9 . David Thomas, Norm Breslow, and Jon at Thorong La (1989) involved in an effort to set up a nuclear-powered surveillance device on topof Nanda Devi, near the Indian border with China, in 1965. The goal onthe 1996 trip was to climb a popular “trekking peak”, Mera Peak (6476m /21246ft) to the south of Mount Everest in the Khumbu region of Nepal. Toacclimate gradually we started the trek in Phaplu; the first few days wereextremely wet with repeated close encounters with leeches and the wettesttenting conditions I have ever experienced. As we got over the pass to theeast of the Duhd-khosi river things started to dry out. We ended up notquite making it all the way to the top of Mera, but thanks to an extremelyfit pair of young porters who stomped uphill through new snow for hours,we did get quite high.The third trip, in 1999, was with Norm, David Thomas, and Vera, and tookus on a three week trip into far less traveled country in Dolpo in western imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018 M. BANERJEE AND R. J. SAMWORTH
Fig 10 . Norm, Vera, Jon, and David up in the Himalayas at the Kagmara La (1999)
Nepal. In the course of those three weeks we went over three high passes andspent time near Lake Phoksundo where Vera fell in love with a young girl(Sangmu Royaka) whom we ended up supporting through school in Dunaiand later in Kathmandu. Quite a magical trip all in all.
Mouli:
I don’t know if you remember, but you and Vera and I were on the sameflight from Seattle to Tokyo on that 1999 trip of yours, by coincidence! Iwas going to India.
Mouli:
Have you hiked in the Alps? Any other mountain ranges?
Jon:
Yes, a bit. But most of my climbing has been in North America: the Cas-cades and Tetons in the US, and the British Columbia Coast Mountains inthe vicinity of Mt. Waddington in Canada.
Richard:
Do you still pursue mountain-climbing a bit? And skiing?
Jon:
Mouli:
What other interests and hobbies do you have?
Jon:
Vera is getting me back into photography. The new digital cameras havephenomenal capabilities, and perhaps I can still learn how to program oneof these gadgets!
Richard:
Thanks, Jon, for that fascinating insight into your life and career. It’sclear you’ve led an active life, both professionally and personally, with moreto come on both fronts in the future. imsart-sts ver. 2014/10/16 file: JAWInterviewArxiv.tex date: August 16, 2018
NTERVIEW WITH JON A. WELLNER Mouli:
Couldn’t agree more with Richard! Thanks for letting us interview you,Jon, and all our very best for the coming years!
References: • Chernozhukov, V., Galichon, A., Hallin, M., and Henry, M. (2017). Monge-Kantorovich depth, quantiles, ranks, and signs.
Ann. Statist. , 223 - 256. • Koltchinskii, Vladimir; Nickl, Richard; van de Geer, Sara; Wellner, Jon A.(2016). The mathematical work of Evarist Gin´e.
Stochastic Process. Appl. , 3607 - 3622. • Chernoff, H. (1964). Estimation of the mode.
Ann. Inst. Statist. Math. ,31 - 41. • Gill, R. D. (1989). “Non- and semiparametric maximum likelihood estima-tors and the von Mises method (Part I)”.
Scand. J. Statist. , 97 - 128. • American Dhaulagiri Expedition - 1969.
American Alpine Journal , Ameri-can Alpine Club.
17 (1) , 19. 1970. • Gardner, R. J. (2002). The Brunn-Minkowski inequality.
Bull. Amer. Math.Soc. , 355 - 405. • Groeneboom, P. (1987). Asymptotics for interval censored observations.Report 87-18. Department of Mathematics, University of Amsterdam. • Groeneboom, P. (1989). Brownian motion with a parabolic drift and Airyfunctions,
Probab. Theory Related Fields , 79 - 109. • Groeneboom, P., Lalley, S., Temme, N. (2015). Chernoff’s distribution anddifferential equations of parabolic and Airy type.
J. Math. Anal. Appl. ,1804 - 1824.,1804 - 1824.