Fall 2020 Statistics/ Data Science PhD application evaluation

[regressionist]

Hi!

I'm a recent graduate planning on applying to PhD programs in statistics or data science for Fall 2020. My research interest lies in the field of Natural Language Processing. 

Undergraduate Institution: Western-style university in Central Asia (Nazarbayev University)

Majors: Mathematics, minor in Economics

GPA: 3.57/4.0 (lower in the first year when my major was Chemical Engineering, and higher after I transferred to Math)

Type of Student: White Female 

Relevant Courses:

Math & Statistics: 

• Calculus 1, 2, 3 (A, A, A)
• Linear Algebra (A)
• Probability (B)
• Applied Statistical Methods (A)
• Math Statistics (B)
• Regression Analysis (A-) 
• Design of Experiments (A-)
• Intro to Proofs (B+)
• Real Analysis (B+)
• Nonlinear Optimization (A) 
• Actuarial Math (A-)

Computer Science:

• Programming for Scientists and Engineers (A)
• Performance and Data Structures (B+)
• Statistical Programming (A)

Other:

• Econometrics (A)
• Economics of Financial Markets (B)
• Capstone Project in Math (A)

GRE General: Q 167 (91%) , V 152 (56%), AW 3.5 (41%)

Research Experience:

My research was generally focused on the development of computationally efficient and interpretable algorithms for obtaining word embeddings and sentence embeddings.

• First publication: Springer "Lecture Notes in Computer Science", topic: NLP (word embeddings),  first author, conference: CICLing 2019, top 10% of papers 
• Second paper: topic: NLP (sentence embeddings), submitted to AAAI 2020
• Research internship in South Korea (UNIST), topic: networks theory 
• In plans: research internship at KAUST
• 1,5 years experience of working as RA 

Work Experience:

• Currently working as a Data Scientist in a Venture company 

Letters of Recommendation:

• One from the research advisor (I also took 3 classes from him) 
• One from an academic advisor (He is currently working at the University of Minnesota)
• One from a professor of Nonlinear Optimization (previously he worked at University of British Columbia)

Concerns:

This year I was accepted to NYU MS in Data Science and UIUC MS in Statistics, but to the several financial circumstances, I was not able to join any for Fall 2019. That is why I changed my plan, and now I want to apply to the PhD program which could possibly provide the assistantship. I want to continue my research in the field of NLP, but after a search, I realized that most of the professors interested in NLP are working in CS departments (or even in linguistics), but my background is mostly related to Statistics. My dream university is NYU (Courant), but I am not sure about my chances. I will be happy to read your advice and grad schools suggestions. Thank you very much!

Edited September 5, 2019 by regressionist

[omicrontrabb]

If you’re primarily interested in natural language processing, you probably shouldn’t get a PhD in statistics. That’s typically a topic in computer science or computational linguistics departments. Many statistics departments have no one working on it.

I’ve done some work on natural language processing and most of the cutting edge stuff, like GPT2 or BERT, is close to pure computer science. Sure there are probabilities in those models, but they don’t mean anything in the traditional statistical sense.

You should probably apply for computer science PhD programs if you want to be an NLP researcher. If you’re more interested in application, I’d focus on gaining experience. A lot of people working in NLP seem to have CS backgrounds and then learn NLP themselves from free online resources.

[regressionist]

@omicrontrabb Thank you for your advice! 

Yes, I also thought about MS or PhD in CS, but the problem is that I don’t have relevant courses in my bachelor transcript - they are mostly from math/econ. On the other hand, my research experience is mostly related to cs - maybe this factor could possibly overweight the lack of cs coursework. I have covered several online courses- but not sure whether they could be counted by the grad admission. 

That is why I have some kind of mess in my head - not sure which program to choose in order to be accepted to PhD. PhD in data science could be ideal for me - but it is offered only in the limited number of universities, such as NYU, Columbia- and the chances of admission are very low given my unknown university and relatively low gpa.

[Stat Assistant Professor]

Your chances of admission for the top 60 or so Statistics programs in the U.S. are slim, since your university is relatively unknown and your GPA is not great. Competition from international Asian students is fierce. As an example, at my program (ranked ~40 by USNWR), most of the students were from the top universities in China, South Korea, or India. At my program, there were a few students from places like Bangladesh and Iran, but they had to be the *very*, very top students in that case (Summa Cum Laude, ranked 6th on the national entrance exam for Masters programs in statistics in their home countries -- stuff like that). To improve your chances, you would probably need to obtain a Masters from an American university and earn all A's there (then you *might* be able to get into a PhD program, but even then, it would have be a mid-tier or low-ranked one). But that would entail going into a lot of debt, which seems to be an issue for you.

I think you have a better chance at Computer Science admissions with your current profile than Statistics. PhD admisisons committees for CS care a *lot* about research experience and usually are more forgiving about lower GPA's (unless it's sub-3.4, they will probably overlook this if you have strong research -- and even when it is sub-3.4, they are likely to forgive it if your research profile is very strong). It seems like you have a decent amount of CS research, since you have a first author paper that ranked in the top 10% of a conference. As @omicrontrabb pointed out, your stated interests also seem to be more aligned with computer science than statistics. There was actually one student in my PhD program who ended up transferred to a Computer Science PhD program because he discovered that Statistics (particularly all the statistical and probability theory we had to learn) was not at all what he was interested in.

Edited September 6, 2019 by Stat PhD Now Postdoc

[regressionist]

@Stat PhD Now Postdoc Thank you for your reply! 

There are a few notes I want to add about my background. Regarding my university- it is ranked top 1 in my country with great international faculty, and our alumni are currently doing PhD in such universities as Stanford, MIT, Columbia, Brown etc. 

My own experience showed that this year MS in Statistics admission seemed not so competitive - students from my class were accepted to UIUC, Columbia, Duke, Purdue - even with lower GPA than mine (3.2-3.4). I also was accepted to NYU MS in Data Science (acceptance rate <10%). Based on this fact I decided that the MS admission is not so hard - the only concern is funding. This makes me think that for me it is easier to got accepted to the Statistics program (I do not aim to the top uni such as Ivy) rather than CS, but may be my hypothesis is false. I don't familiar with PhD admission, but from your words, it seems that it differs very much from the MS admission.  If PhD admission priorities GPA and advanced undergraduate coursework such as Real Analysis, Abstract Algebra etc - then my chances are very low 

I am not really sure whether I actually want to be a statistician, CS major is more interesting to me - but I do not have relevant courses in my transcript (such as Operating Systems, Computer Networks etc), and this makes my transition from math to CS harder. Not sure what to do now.. 

Edited September 6, 2019 by regressionist

[Stat Assistant Professor]

1 hour ago, regressionist said:

@Stat PhD Now Postdoc Thank you for your reply! 

There are a few notes I want to add about my background. Regarding my university- it is ranked top 1 in my country with great international faculty, and our alumni are currently doing PhD in such universities as Stanford, MIT, Columbia, Brown etc. 

My own experience showed that this year MS in Statistics admission seemed not so competitive - students from my class were accepted to UIUC, Columbia, Duke, Purdue - even with lower GPA than mine (3.2-3.4). I also was accepted to NYU MS in Data Science (acceptance rate <10%). Based on this fact I decided that the MS admission is not so hard - the only concern is funding. This makes me think that for me it is easier to got accepted to the Statistics program (I do not aim to the top uni such as Ivy) rather than CS, but may be my hypothesis is false. I don't familiar with PhD admission, but from your words, it seems that it differs very much from the MS admission.  If PhD admission priorities GPA and advanced undergraduate coursework such as Real Analysis, Abstract Algebra etc - then my chances are very low 

I am not really sure whether I actually want to be a statistician, CS major is more interesting to me - but I do not have relevant courses in my transcript (such as Operating Systems, Computer Networks etc), and this makes my transition from math to CS harder. Not sure what to do now.. 

Yes, PhD admissions in Statistics will prioritize performance in advanced undergraduate math classes, and the PhD programs themselves are more mathematical n nature (at least for the coursework portion in the first two years, which is very much of the "theorem/proof" variety). Masters admissions is in general not competitive -- as long as you have decent grades in Calculus I-III and Linear Algebra and a decent GRE Q score, you should be able to get into some program.

I'm not sure if your undergrad degree is that important for CS PhD admissions. I have seen people with biology and linguistics degrees get admitted to PhD programs in computer science, provided they have relevant research experience that aligns with their PhD advisor's lab. It seems like you do seem to have relevant research experience, having published in a paper in a conference that was in the top 10% and having submitted a paper to another conference. I would do research on professors in Computer Science departments who match your research interests and reach out to them to see if they are accepting PhD students. Admissions decisions in CS are often made in accordance with lab PI's who accept new students into their labs.

Read this article from Philip Guo on how admissions decisions are made in Computer Science, and note the sentences: "the property that best separates good from bad applications is research density," and "GPA: warning sign if too low, but usually don't care. It's rare that someone with strong research credentials has a dangerously low GPA, and even if that were the case, I wouldn't care much." http://www.pgbovine.net/PhD-application-tips.htm

 

Edited September 6, 2019 by Stat PhD Now Postdoc