Choosing Statistics PhD: Harvard vs Berkeley?

[Ryuk]

I have been admitted to the statistics departments at both Harvard and Berkeley. I applied to 24 schools, so I've also been accepted to virtually all of the top 20 stats programs, excluding Stanford.

My academic interests are pretty broad, but I'd like my research to be more theoretical and in the realm of probability or machine learning/deep learning, if possible. I'm also not sure if I will try to go into academia or into a research team at Google, Microsoft, Facebook, etc.

I am mainly considering these two because Berkeley is so good at ML, but Harvard is a better fit in every other way (culture, location, etc.). Any advice would be appreciated! I am also happy to provide any more information.

 

 

 

[MathStat]

how is harvard a good fit given your research interests? I feel like they're more into biostat/applied stuff..although Cynthia Dwork is there...

If you're into probability, deep learning etc, then berkeley and potentially other schools out of those 24 would be better fits. 

 

 

[Ryuk]

@MathStat

Thanks for the reply!

I know that Berkeley is a better fit for my research interests and I know that research fit is one of the most important parts of choosing a PhD.

However, I also hear people say that you should choose a place that you would enjoy and fit in for the next 5 years. I think I would enjoy my time at Harvard more. In addition, Harvard grads seem to do extremely well on the job market.

There are young faculty at Harvard with similar research interests as me, but obviously there are many more options at Berkeley. I am having difficulty comparing my desire to live in Boston and be a part of the Harvard community with the exact research fit at Berkeley.

[MathStat]

Are Harvard grads getting jobs at Google, Microsoft, Facebook, etc Research? I have not been stalking recently lol, but I cannot recall any examples (I wouldn't mind seeing some if you know any!). And the ones who get the top academic placements (aka berkeley and stanford TT professorships) seem to have worked in causal inference. If that is a strong area of your interests, then sure, go ahead.

I turned down Harvard (and had similar interests to yours and even a similar situation, haha), cause I thought there were only 1-2 people with similar interests as mine.

another bit of advice when you have to decide between almost all the top programs is to not get hung up on the top 2 ones that are ranked just after stanford. I think other ones you should also consider carefully given your interests are Columbia, UPenn Wharton, Duke, Yale (only if you wanna do pure math stat; imo they're some of the best at that). Make sure you review these carefully as well. 

Edited February 18, 2021 by MathStat

[Ryuk]

@MathStat 

You're right that only a few people at Harvard do what I'm interested in. I guess I'm wondering how much I should force it if it's a place I really want to be. 

A quick glance at your profile seems to tell me that you are currently at U Chicago. I also have an offer from them. If that's true and you really were in a similar situation as me, why did you pick Chicago over Berkeley and Harvard?

[MathStat]

Chicago over Harvard was a no-brainer given my research interests. Of course that Harvard is still an outstanding program, and turning it down is not easy either way. 

Chicago over Berkeley was a hard and excruciating decision, but I reasoned that there were only a handful of Berkeley professors I truly wanted to work with all of which are absolute superstars and who have millions of students and postdocs...Chicago was an equally good match for my interests, they have TTIC which is at least top 5 for theoretical machine learning, as well as a handful of superstars or rising stars in the Statistics department and the Booth school of business. Also, besides purely academic reasons, between very long winters and very high living costs, I decided I'd prefer the former, haha. But this is a purely personal preference, and I totally understand people who think otherwise. 

You should also consider Chicago and should come at the visit! We do have the notorious quals, as well as the coursework which take up all of the first year plus summer. This is not for everyone. Research-wise, I am still recovering from that, yet I do have two exciting lines of research going on...I guess I just need a few more months to be able to tell you exactly how it's gonna turn out. 

[Stat Assistant Professor]

Yeah, Harvard is really, really strong in the areas of causal inference and MCMC. For deep/maching learning and probability theory, I would say that Columbia, UC Berkeley, and UPenn Wharton have an edge over Harvard (e.g. you've got David Blei at Columbia, Michael Jordan and Martin Wainwright at Berkeley, Edgar Dobriban and Weijie Su at UPenn, etc.). There is also a large group of probability theory researchers in the Statistics Department at UCB, which is somewhat unusual nowadays (typically there is only one or two faculty in a Stats department working on pure probability theory topics).

Edited February 19, 2021 by Stat Assistant Professor

[bayessays]

How sure are you of those research interests and how passionate about them are you?  Some people can be truly fulfilled by their research and if that'll make you happy, go to Berkeley.  But you're not even going to be able to do good research if you're unhappy and wishing you were on the other side of the country.  Are you sure that you are that much interested in probability than say, MCMC, where you could work with Xiao Li Meng at Harvard who to me is one of the most interesting people in statistics - just read some of his paper titles and listen to his talks.  Are you sure that theoretical machine learning at Berkeley is that much more interesting to you than the reinforcement learning that Susan Murphy is doing?  There's plenty of theoretical stuff going on at Harvard that might satisfy you intellectually, and I definitely think that location is extremely important.  The facts are that you will be qualified for top stats jobs after working with someone good at Harvard.  Maybe Berkeley will offer you a slightly better chance at doing the type of ML that gets a FB research job, but is that extra slight chance worth 5 years?

My recommendation would be to download some papers from profs you like at both school.  Read the papers from Berkeley and ask yourself if you love reading about that subject so much that you would move across the country to Berkeley to be able to ask the person who wrote it a couple questions every week.

[DanielWarlock]

I'm a current student at Harvard and was also admitted to Berkeley last year. I know exactly what you are talking about regarding "hot areas". Our department has been recruiting "superstars" in the "hot areas" such as machine learning, high-dim, network, deep learning, non-convex optimization etc. There is one more assistant prof coming in next year working in one of these areas. I think Harvard is becoming less Bayesian/MCMC with the new recruits. A thing to note is that most people in our department working in those areas are "younger generation" who are generally students of well-established people found at, say, Stanford. 

That said, a factor you should not ignore is access to MIT. It's possible to find advisors at MIT who are working in the "hot areas",  Rakhlin, Poggio, Rigollet, Moitra, Mossell, Jaakkola, to name a few. Similarly, you can find some very famous people in random matrix, statistical physics (e.g. Yau) in harvard math department as well, who actually supervises student doing statistics type of work. In short, I don't think you are losing much academically by choosing Harvard over Berkeley.

I don't think Berkeley's job prospect is any dimmer than Harvard (if not brighter). The only exception is if you are going to a foreign country working in something unrelated to statistics/CS (e.g. banking), then Berkeley probably sounds less impressive. Some HR in my country even thinks of Berkeley as some sort of cash-cow spin-off of "University of California". So you could even get questioned on this. 

[MathStat]

Is MIT access really feasible as a harvard stats student? If so then yup, huge point taken. 

[Ryuk]

@DanielWarlock Thanks for the input! My main concern is it seems that Berkeley has a strong reputation in ML and a close relationship with the CS department, while attending Harvard would require me to take a bit of a gamble. If I want to do anything with deep learning or theoretical machine learning, I will have to reach out to the CS department, which doesn't seem to have a close relationship with the stats department. I will also be counting on some of the new hires to be good research fits. I am definitely attracted to Harvard, but the reasons I just listed make me nervous about research there. To what extent do you agree with my analysis?

@MathStat I spoke with a few current students who said that there is one Harvard PhD student with a co-advisor at MIT, but he also got his masters from MIT. I'm not sure how feasible it is for the average student.

[MLE]

Some professors at Berkeley also have ties to MIT through FODSI which has been very active this year. I imagine these ties will get stronger over time with FODSI postdocs at Berkeley. This tie is probably not as strong as Harvard, but it's worth noting that it is there.

[DanielWarlock]

10 hours ago, Ryuk said:

@DanielWarlock Thanks for the input! My main concern is it seems that Berkeley has a strong reputation in ML and a close relationship with the CS department, while attending Harvard would require me to take a bit of a gamble. If I want to do anything with deep learning or theoretical machine learning, I will have to reach out to the CS department, which doesn't seem to have a close relationship with the stats department. I will also be counting on some of the new hires to be good research fits. I am definitely attracted to Harvard, but the reasons I just listed make me nervous about research there. To what extent do you agree with my analysis?

@MathStat I spoke with a few current students who said that there is one Harvard PhD student with a co-advisor at MIT, but he also got his masters from MIT. I'm not sure how feasible it is for the average student.

I can't say any professor (MIT or Harvard) will agree to take you as student. Same thing with Prof Jordan or Wainwright at Berkeley. I talked to one of their students on admit day who told me they are extremely busy. But it is not as hard as you think to approach CS or MIT professor at Harvard. Why is there no close relationship to stats department? Jansan, Murphy, Ba have cross appointment. Yue Lu held a reading group last semester with two stats faculties last semester. I found a paid research with CS prof last summer on graphical neural networks without knowing much about the subject (although I ended up not taking it due to COVID). I have not yet approached MIT professors to ask them to advise me on an official capacity but I think it is definitely within reach especially if you know their stuff. It is very easy to initiate a conversation and get to know people on personal level. For example, Moitra held a grad seminar (joint with Boaz from harvard) this semester which essentially requires you to talk to him about research on deep nn. I'm taking a class with Rakhlin and Rigollet on high-dim stats and can go to talk to these guys on office hour every week. I feel that most people are very approachable in the first place. The issue is how well you can convince these professors to advise you on official capacity. There is no guarantee with this kind of thing even if you go to Berkeley. 

[Stat Phd]

I would say Berkeley is far way superior than Harvard, this would be a no brainer to me unless you have strong location preferences.

Edited March 6, 2021 by Stat Phd