Qualifications for PhD Programs

Question

Yann LeCun mentioned in his AMA that he considers having a PhD very important in order to get a job at a top company.

I have a masters in statistics and my undergrad was in economics and applied math, but I am now looking into ML PhD programs. Most programs say there are no absolutely necessary CS courses; however I tend to think most accepted students have at least a very strong CS background. I am currently working as a data scientist/statistician but my company will pay for courses. Should I take some intro software engineering courses at my local University to make myself a stronger candidate? What other advice you have for someone applying to PhD programs from outside the CS field?

edit: I have taken a few MOOCs (Machine Learning, Recommender Systems, NLP) and code R/python on a daily basis. I have a lot of coding experience with statistical languages and implement ML algorithms daily. I am more concerned with things that I can put on applications.

Answer 1

If I were you I would take a MOOC or two (e.g., Algorithms, Part I, Algorithms, Part II, Functional Programming Principles in Scala), a good book on data structures and algorithms, then just code as much as possible. You could implement some statistics or ML algorithms, for example; that would be good practice for you and useful to the community.

For a PhD program, however, I would also make sure I were familiar with the type of maths they use. If you want to see what it's like at the deep end, browse the papers at the JMLR. That will let you calibrate yourself in regards to theory; can you sort of follow the maths?

Oh, and you don't need a PhD to work at top companies, unless you want to join research departments like his. But then you'll spend more time doing development, and you'll need good coding skills...

Answer 2

Your time would probably be better spent on Kaggle than in a PhD program. When you read the stories by winners (Kaggle blog) you'll see that it takes a large amount of practice and the winners are not just experts of one single method.

On the other hand, being active and having a plan in a PhD program can get you connections that you otherwise would probably not get.

I guess the real question is for you - what are the reasons for wanting a job at a top company?

Answer 3

You already have a Masters in Statistics, which is great! In general, I'd suggest to people to take as much statistics as they can, especially Bayesian Data Analysis.

Depending on what you want to do with your PhD, you would benefit from foundational courses in the discipline(s) in your application area. You already have Economics but if you want to do Data Science on social behavior, then courses in Sociology would be valuable. If you want to work in fraud prevention, then a courses in banking and financial transactions would be good. If you want to work in information security, then taking a few security courses would be good.

There are people who argue that it's not valuable for Data Scientists to spend time on courses in sociology or other disciplines. But consider the recent case of the Google Flu Trends project. In this article their methods were strongly criticized for making avoidable mistakes. The critics call it "Big Data hubris".

There's another reason for building strength in social science disciplines: personal competitive advantage. With the rush of academic degree programs, certificate programs, and MOOCs, there is a mad rush of students into the Data Science field. Most will come out with capabilities for core Machine Learning methods and tools. PhD graduates will have more depth and more theoretical knowledge, but they are all competing for the same sorts of jobs, delivering the same sorts of value. With this flood of graduates, I expect that they won't be able to command premium salaries.

But if you can differentiate yourself with a combination of formal education and practical experience in a particular domain and application area, then you should be able to set yourself apart from the crowd.

(Context: I'm in a PhD program in Computational Social Science, which has a heavy focus on modeling, evolutionary computation, and social science disciplines, and less emphasis on ML and other empirical data analysis topics).

Answer 4

I am glad you also found Yann LeCun's AMA page, it's very useful.

Here are my opinions
Q: Should I take some intro software engineering courses at my local University to make myself a stronger candidate?
A: No, you need to take more math courses. It's not the applied stuff that's hard, it's the theory stuff. I don't know what your school offers. Take theoretical math courses, along with some computer science courses.

Q:What other advice you have for someone applying to PhD programs from outside the CS field?
A: How closely related are you looking for. Without a specific question, it's hard to give a specific answer.

Answer 5

You have the option of joining a PhD program in business school and information school as well. There are quantitative professors and data scientists in business schools and information schools as well (About US, I am sure there are a lot of schools). This way you are qualified or even over-qualified in terms of quantitative and technical skills and you can spend your time on reinforcing other skills.