Radev pointed me to this discussion by John Langford of Carnegie-Mellon’s new Machine Learning department. I don’t have much to add to the links and comments posted there, but I’m generally supportive of new academic departments, or else of letting existing departments become more flexible about requirements. I think Columbia’s Statistics Department would improve by splitting into two separate departments:
– Applied Statistics
– Probability and Theoretical Statistics
or else becoming a single department of Statistics and Probability with formal tracks for students and faculty in the different subfields. As it is, the statisticians end up suffering through measure theory and the probabilists spend a lot of time teaching intro statistics to unhappy undergrads. I don’t think there’s anything wrong with statisticians learning measure theory, but in practice it takes valuable time and effort that they could instead be using to learn computer science, or economics, or some other sister discipline. (I agree, perhaps, with John’s statement of “‘rogramming as the missing member of reading, ‘riting, and ‘rithmetic.”)
Anyway, having a Machine Learning department sounds like a good idea if it means that the students and faculty there can have a bit more flexibility in what they can do. I also think of statistics as a branch of engineering but it’s not usually in engineering schools. (Columbia has an operations research department in the engineering school but they also do a lot of theoretical probability; some sort of rejiggering seems possible to me, just as they moved Dallas out of the NFC East (I assume they’ve done that by now???) etc.)
Slightly related is the fact that it can be difficult to persuade statistics Ph.D. students to take courses in experimental design and sample surveys, even though these are huge application areas. And then there’s this. My experience at Berkely taught me to be an intellectual pluralist.
A new Master's programme in Bayesian Statistics of six European universities is currently accepting candidates. The proposal of the programme
http://www.uv.es/valenciameeting
-> BayesMasterProposal
contains a summary of the "Bayesian paradigm" with descriptions of different Bayesian career opportunities, including "Academic statisticians" and "Machine learning specialists" (hopefully not mutually exclusive categories).
The proposal reminds me of another document which describes the Bayesian paradigm: the draft issued by FDA called "The Guidance for the Use of Bayesian Statistics in Medical Device Clinical Trials – Draft Guidance for Industry and FDA Staff". I don't remember if this has been already discussed in the blog but it is interesting that FDA is now accepting Bayesian statistical analyses.
One of the curious points in the FDA draft is the instruction not to be volatile: "…we recommend you do not switch from a frequentist to a Bayesian analysis (or vice versa)". Pick your side!
Some points:
1. CMU is one of 5-6 places with *schools* of Computing (others include Utah, UC Irvine, and Georgia Tech). Being a school, CMU can choose to have departments in any sizeable area of CS. They already have units in Robotics, Software Engineering, Language Technologies, etc.
2. UC Irvine is another such place with a school of Information and Computer Science with three units – Information Science, Computer Science, and Statistics.
3. At Michigan, we have a department of Electrical Engineering and Computer Science under Engineering and separate units in Statistics (in the Literature, Sciences, and the Arts college) and Information Science (separate school).
I personally like Irvine's model best as it brings all related units into one place.
How to learn programming for statisticians?
I can see how this could be useful. My poor efforts at R programming ate a testament that I need to learn to do it better. I've tried a few books, but constantly get frustrated when I go away from the examples or very close approximations to the examples.
To me, "statisticians end up suffering through measure theory" sounds no different from, say, "programming languages people end up suffering through formal logic". How are the students supposed to get an in-depth understanding of the different types of convergence — the very core of statistics — without measure theory?
Leonid,
Although I have never studied measure theory myself, I am happy to agree that it is a useful thing for a statistician to know. Other useful things for a statistician to know, ranging from fundamentals to important topics, include:
– programming
– optimization (including numerical approaches
– statistical graphics
– distribution theory (from the multivariate normal distribution up to more advanced models such as arise in spatial statistics)
– analysis of variance
– sampling and experimental design
– applied topics including economics, biology, etc.
I'm happy for measure theory to be an option, but I don't think it's a good idea to have two required semesters of measure theory and zero required semesters of all of the above.
To get back to your original question, I can accept that convergence is "the very core of statistics" as you understand it, but for many other statisticians, the "core" will be different and might very well be programming, or optimization, or graphics, or …
But for applying statistics "almost surely" nothing more than convergence in distribution needs to be understood …
But part of the problem comes from the habit of those who can teach measure theory of trying to train measure theorists rather than imparting the basic knowledge of measure theory to those who might want to do something else.
(And think will all often make that mistake)
Keith