icantdoalgebra

Last year I've asked for an extension on April 15th due to getting an offer very late in the cycle. IMO it doesn't hurt to try and ask at least; the most likely worst-case scenario is that your request for extension gets denied (I would be quite shocked if anything worse than this happens) but there is a chance that they would grant you an extension.

As a side note to this point, a lot of Berkeley's most famous statisticians have passed away. Note this is not particularly relevant to the current discussion at hand since we are discussing the departments today (and also because ... dead people don't do research), but if we do want to talk about "revolutionary" developments I think it is somewhat worthwhile to bring up Berkeley's significance (although it really shouldn't affect anybody's decision in where to go or departmental rankings because this is quite historical). Jerzy Neyman was a professor at Berkeley and was the key driving force behind the formation of the department and he was no slouch when it came to revolutionary developments (quotes deliberately omitted because confidence intervals, hypothesis testing, potential outcomes are all, beyond any doubt, revolutionary developments). George Dantzig was a student and professor at Berkeley for a while (although he left for Stanford later) but while at Berkeley he developed the simplex algorithm which may not be revolutionary in statistics, but is the foundation for several other fields. David Blackwell was a professor at Berkeley for several decades; you might know him from Rao-Blackwell, but he also did a lot more than that (Blackwell-Macqueen, etc.). There are a few others that can be mentioned too, David Freedman has had quite a bit of influence outside the world of statistics (and in statistical education), Leo Breiman has been mentioned above (CART and random forests), Le Cam and his work on lower bounds, and several more. I bring them up not because I'm trying to argue against any of the positions taken here: I agree that Stanford's statistics department is better today. It is in a class of its own when you consider traditional statistics. But I still feel somewhat obligated to defend Berkeley if we are mentioning "revolutionary" developments without mentioning Berkeley's own rich history. However let me re-emphasize, none of the above should be a part of one's consideration for picking a program for grad school nor should it be a part of decision making about which departments are stronger.

Easy; Peter Bickel, Michael Jordan, Martin Wainwright. People like van der Laan, Bartlett, Brillinger, Aldous, Yu, and Pitman might not be as famous as those Stanford faculty that you listed, but its not like they are some random professors. How about Fernando Perez and his work on creating Jupyter Notebooks? Perhaps its not as much of a research accomplishment, but creating a widely-used software that makes performing statistics and data science easier is definitely worth quite a bit of influence. I'm not trying to argue that Berkeley is better, in fact I agree that Stanford's program is better. However is it substantially better? I am skeptical. There is a valid point though, in that Berkeley's stats department has leaned more heavily towards machine learning recently and if you are a statistics purist you could reasonably make the argument that Stanford is substantially better if that is your criteria for evaluation.

If any stats program decided to do this, you would instantly know to avoid it because this is, by definition, extremely horrible statistics. The goal of an admissions committee is to try and pick the candidates that has the highest future success (of which is measured to whatever the committee decides to chose) subject to a series of resource constraints. Ultimately there is a regression problem involved: trying to predict future outcomes given the applicant profile. You are proposing for the committee to throw out all covariates that may have predictive power on the outcome and replace it with a single measured value (the proposed test score), whereas any statistician with half a brain would propose the exact opposite. The admissions committee may not be running a linear regression or some ML algorithm to try and predict what will happen to candidates in the future, but the point still stands. Having more than just one feature to look at is often superior to only looking at one feature and trying to predict with that.

Having a award itself doesn't mean much, but usually stronger candidates tend to have more awards. As an aside, you should just be focusing on doing good research with your current professor; all the professors in the stats department at Berkeley are highly qualified (that's what it means to be a top department) so even if they aren't Lebron James, a strong letter of recommendation from them will carry a lot of weight amongst admission committees. Fixating on the fame of the professor might turn out to be quite counter-productive to your current research. As per your point of delaying graduation, I really don't know what the optimal move is. I don't think having good grades in 210A/B will be that much more valuable than just having taken 210A because the grading in grad classes is quite generous. Having another research opportunity to replace a recommendation letter which basically states you did well in a class would be helpful but is it worth delaying graduation for? I have no clue.

Berkeley last year did a batch in February and a waitlist batch in April. Rejections usually come a few weeks later in the cycle. Correction: Berkeley apparently doesn't officially do waitlists; usually to admit more people they have to get approval.

You should be a strong candidate for a place like Duke, but for Berkeley and Stanford I'd say you have a decent chance but I wouldn't bet on it (but I wouldn't bet against though). I was class of 2020 at Cal and it seems like your profile is somewhat weaker than some of my friends, both in terms of coursework and research; none of us got into Stanford or (I think) Columbia, and very few of us got into Berkeley. However this assessment is based on your current position, and a lot can change in the span of a year. As for your question about research publications, I'd say the actual research you do matters more than any publication. People with publications tend to better in admissions not because of publications, but because usually those with publications have stronger relationships with the professors writing them recommendation letters. I had zero publications when applying to grad school and I got into places, as did some of my friends. About delaying applying to a year later, I had asked essentially the same question to a lot of people, and the response I got was mostly "it really doesn't matter". It doesn't seem like they have higher requirements for those who apply the year after graduating but its not like you can do much to strengthen your application further with the exception of finding another research opportunity.

Stats PhD.

I didn't take multivariable calculus when applying last year and didn't have trouble getting into programs. I think having more advanced classes should make up for not having an intro class like multivariate, but that's just my guess and maybe you should take it to be on the safe side? I'm not too certain what the correct move is.

So most of the programs I applied to allows for late submission of letters, especially those with December 1st deadlines. As for whether it has an impact or not, it's hard to say. I've had a professor tell me that if it comes late for December 1st deadlines it doesn't really matter because nobody wants to look at applications seriously before the New Year. As for personal anecdotes: my rec letter came in late for the early December deadlines, and I got into one out of three programs but this is very low signal information since its a sample size of N=3 and the difficulty of admission is quite different in this group.

Isn't that more due to the fact that MIT primarily only has people in either ML or high-dimensional statistics (and in this case I feel like the opportunity of working with Candes at Stanford or Wainwright at Berkeley is probably better than being at MIT), but there's more to statistics than these two fields: what if the person wanted to do something in causal inference or post-selective inference? MIT would clearly be not as great compared to other top stats programs. Additionally what makes a good EECS application is pretty different from what makes a good stats application: applying to EECS in ML usually revolves around having published somewhere in ICML or NeurIPS (sometimes even multiple times) as an undergrad, and comparatively grades/coursework really don't matter that much. Perhaps this is true in general about stats program curricula being outdated but if we are comparing Stanford/Berkeley to MIT this isn't really that true nowadays. People in the stat department who are interested in ML are encouraged to go take courses in optimization and information theory (which is what I am doing) because as you said, they are very useful. In fact one of the core courses at Berkeley gives a brief introduction to information theory because of how useful it is. But again, not everybody wants to do ML/data science/what-ever gets hyped up in the media nowadays and places like MIT EECS/Math or Princeton ORFE are really niche recommendations since they tend to lean very heavily towards a small subset of areas within statistics and have virtually no presence elsewhere. Also about your point for professorship, I know Berkeley gave out 3 offers for faculty last year, and 2 of them are for somebody with a traditional statistics background. @kimmy sorry for the sidetracking but the people on this forum are pretty good at giving advice on applying to stats programs (probably because there are a few stats professors running around on these forums). I'd be somewhat more skeptical of any advice you'd receive here about applying to math or EECS (as to the best of my knowledge) because there aren't any similar such people on this forum and you should probably go look elsewhere for advice but do consider these options if that is something you are interested about.

As recommended go check out the mathematicsgre forum, there's a lot more people there involved in math. But I will say that I know many people with applications that are quite a bit stronger than yours (people at schools like Berkeley, Stanford, and Harvard) with higher gpas, 5+ graduate courses (some even reaching double digits!), multiple high-quality REUs, GRE scores in the 90+ percentiles who weren't able to crack programs like Princeton. Your recommendations seems like a bit of a wildcard but I can imagine a situation where if your profs call you brilliant in the rec letters you may get in. If you have some sort of story about overcoming adversity with your low CC GPA and turning it around when you transferred, manage to get a pretty high score on the subject test I don't think Berkeley is an impossibility, but still a reasonable reach. You should take all my advice with a grain of salt though as math is not my area.

I think you should clarify if its a school known for grade deflation or not. If its UChicago or Caltech the response might be different than if it were say Harvard or Yale. I would be overstepping my expertise if I tried to tell you what schools are a target vs reach but I could imagine the answers changing. Maybe others can comment on this perspective.

You still have roughly 5-6 months before you need to apply; you can find a research opportunity in statistics at least by the time the fall semester starts, and doing a semester of stats research would get you a pretty strong rec letter. It shouldn't be that difficult to do considering your grades and the fact that you are at an Ivy league school. Usually domestic students don't have meaningful theoretical research (although this is becoming less and less true) before applying to a PhD program, but international students at the top programs often do (to be fair I've only interacted with those at Berkeley but I'd assume it be similar at peer institutions). If you think that taking a gap year means that you get to apply later and have an easier time getting in due to Covid being over, I'd say just apply this year; there's really no guarantee things will be better. If you can do something in the gap year that would greatly improve your application, like a serious theoretical statistics research opportunity, its not a bad idea.

I would second the Macbook if you aren't that tech-savvy, as having a Unix based OS is much easier to deal with than a Windows one for anything programming related. If you don't like Apple, then an option is to dual boot Linux on whatever Windows machine you buy. For anything that requires intensive compute, your university will likely either have a cluster or provide you with cloud compute credits, in which case you would need to ssh into and having a Unix OS makes this process significantly less painful. The Windows Linux support IMO is frankly quite awful and package management/installing new programming libraries is also horrendous on Windows. Also second the insanely expensive graphics card, think about how fast your ggplots will render with real time ray tracing.

As an aside I think there are benefits to applying later in the cycle (especially as late as possible) if they don't mention that they do any sort of rolling admissions, as you can take the fall semester to improve your relationship with the professors writing your rec letters. I say this because I managed to get a research opportunity late August and the recommendation letter I got from that definitely strengthened my application profile considerably, but it doesn't have to be research. You could try and initiate a DRP with a professor writing you a rec letter or be a TA for one of their courses.

210B just requires some basic martingale stuff from 205A (sometimes not even). I took 210B without 205A/B and I was fine. The reason I recommended 210B is that you can get 205 anywhere: every measure theoretic probability theory course will cover basically the same material. However 210B is really a Berkeley specific course and you would have a hard time finding a similar course elsewhere; for example Berkeley's Stats is really big on non-asymptotic results and 210B introduces a lot of the tools and techniques for non-asymptotic results so you would get a decent idea of some of the research approach that people in Berkeley specialize in. I think that if you have an opportunity to take this instead of 205, which you can just take the equivalent at any university you go to, you should.

Agree with most of what's been posted above before: you should probably be thinking about applying to PhD programs rather than masters. I'm currently a senior at Cal and you have a pretty similar profile with me and I got into several top PhD programs in statistics. However I'd be a bit careful this coming application season since of the Covid-19 induced recession, it seems like schools are planning for smaller cohort sizes (a rumor that I've been hearing going around but unconfirmed) so it might be tougher to get into top programs compared to last applicant season. I'd swap out 205 for 210 though; primarily because the content you see in 205 you will be able to see anywhere else, but the content covered in 210B is very unique to Berkeley and I think you really would be missing out if you didn't take advantage of the opportunity to take it while you are still at Cal. PM me if you have any further questions.

Actually the criteria for getting into ML are changing somewhat, or at least for Berkeley. Part of the reason why the publishing requirement is so high to get into PhDs is that the bar for publishing in a top conference like ICML or NeurIPs is far lower than publishing a paper in a top stats journal like AoS, JASA, or JRSS-B. ML has this reputation of being a field where you just have really large labs and just churn out as many papers as you can every year (Levine at Berkeley who does RL research submitted 40+ papers to a conference on RL and had I think 24 of them accepted). This creates a issue with the people who review the papers, as there simply is too many to review each one thoroughly. Berkeley to an extent has recognized this problem (one professor even told me that conferences in ML are filled with "garbage" papers) so they now value letter of recommendations far more than publishing record. But again, usually the best letters of recommendations comes from Professors who you've done research with. As for Stats vs ML, I think you should apply to both programs where you can. For example, CMU lets you apply to multiple programs so there's really no reason not to do so (unless the financial requirement creates undue hardship). Even if you have to choose, a lot of schools have faculty in multiple departments because they recognize that the work is very similar and you can work with these faculty members regardless of department. Although it also depends on how strongly you want to do ML, as you would be hard pressed to find statistics faculty that aren't joint CS professors who do deep learning; everything else you would be able to find in a statistics department.

Penn is a very small program; I think their cohort size is 5 a year so their department might not be as strong as say, Michigan or CMU, but getting in might be as or more difficult because CMU and Michigan have cohort sizes of about double that. Also in general, Ivies are harder to get into relative to their departmental strength due to their name outside of statistics.

Agree with the above; your profile seems a bit weird to me. Analysis and biostats are really far from each other although you do have another year to figure out roughly what you want (but you shouldn't try and do both) In addition if you are applying to math programs in analysis, (and want to continue doing PDEs), I think UCLA is a pretty egregious omission I personally didn't go through math graduate admissions, but a lot of my friends did so I can say a little bit through that. You should probably apply to a broader range of schools than just the top few: I know people who took 8+ graduate courses at Berkeley (which is, according to USNews, a top 5 math program) with all As and still had a hard time cracking the Princeton/Harvard/MIT/Stanford tier of schools. I think there is also a tendency for math graduate schools to reject their own undergrads, since its usually encouraged for undergrads to go somewhere else for their PhD.

I think real analysis is a good choice, although I think that Rudin is awful textbook to self-study from, especially for the first time. Rudin really shines if you've already learned real analysis and you want to go over it again (and I think many should do this, because mastery doesn't come from learning something once, but reviewing the material over and over again). For a first time introduction to real analysis, and you don't have a really strong math background, I think Ross is the standard textbook. For an honors level introduction to analysis I would highly recommend Pugh since it is filled with pictures that help illustrate many concepts.

Undergrad: UC Berkeley Major: Computer Science GPA: 3.96 (4.00 major gpa) Student Type: Domestic, Asian Male GRE: 168V(98)/170Q(96)/5.5W(98) (percentiles might be off) Math Subject: 840 (84) Courses: Honors Linear Algebra (A+), Honors Abstract Algebra (A), Honors Real Analysis (A-), Honors Complex Analysis (A+), Numerical Analysis (A+), Grad Analysis I and II (A, A), Probability Theory (A+), Mathematical Statistics (A+), Theoretical Statistics I, II (for PhD Students, A, A), Probability for PhD Students (A-), Intro to Programming (A), Data Structures (A), Discrete Math and Probability (A+), Intro Computer Architecture (A), Databases (A), Algorithms (A), AI (A), Machine Learning (A) Programs Applying: Stats PhDs Research Experience: Currently working with a professor and his students on Bayesian nonparametrics, trying to wrap it up and publish soonish (likely second author, contributed some theoretical and numerical results), also working with another professor on more applied machine learning research, writing code to implement it as well as other potential modifications and applying it to more data Work Experience: Spent a summer at a startup doing some natural language processing, another summer at a quantitative asset management firm Letter of Recommendation: Professor doing research with (hopefully would be decent, but met him twice in person in the past 6 months), another professor that I took Theoretical Statistics with (will also hopefully be decent as I think I stood out a decent amount in that class, actively participating and going to office hours), and the third professor is the also the one I'm doing research with (but I've met with them once) Results: Stanford: Waitlisted UC Berkeley: Haven't heard back, talking to some people on the inside, seems like a waitlist Harvard: Rejected UW: Ghosted, likely rejected Chicago: Waitlisted CMU: Accepted Duke: Accepted Michigan: Accepted NCSU: Accepted (Declined) Columbia: Rejected Miscellaneous Points: I think Duke was really interested in the research I did and that probably helped me out a lot. The main boosts to my application are probably taking a lot of hard math classes, as well as the PhD statistics classes at Berkeley, and doing research with some pretty big names in the field. My main failings probably came from the fact that its fairly hard to establish connections with professors as an undergraduate at Cal, and after trying to get involved in research for about 5 semesters straight I only managed to get it the semester when I was applying to PhD programs. Best of luck to anybody reading this from the future who are also applying!

Usually a strong math background means something like a combination of real analysis, linear and abstract algebra as well as more advanced "pure" mathematics classes. However depending on the program "advanced calculus" might be equivalent to real analysis elsewhere. so you'd need to use your judgement on that.

Echoing some of what bayessays has mentioned, their department is on the smaller side in terms of traditional statistics but he also has a high opinion of Michigan assistant professors; he's mentioned that they've made some good hires over the past few years. Also one piece of advice he gave that might be useful to everybody: when choosing a PhD program and considering potential advisors, it shouldn't be about only one person you want to work with. There should be some degree of robustness for an advisor: you should have multiple people at program you're happy to work with in case things don't just work out.