18

I'm a first year postdoc in pure mathematics (geometry/topology with strong background in analysis) with undergraduate statistics and probability background (with also measure theoretic probability). I also have some programming knowledge in ForTran, C and Matlab, but I never used them in my pure mathematics career in my graduate school for the last 5 years.

In my next job, I'm considering doing a postdoc in machine learning (ML). The reason behind this switch is: I'm satisfied with my pure knowledge so far and have been wanting to see some real-life applications of mathematics, and also keep my options open in both industry and academia.

My questions are:

  1. How hard is this switch going to be? I guess I've all the required mathematics background, but will it be hard to pick up the necessary computer science skills, even if I work in more theory-oriented problems?

    What exactly are the programming knowledge I need to master to work in ML?

  2. Is there a website/email-list where I can get notifications on jobs in machine learning? I'm looking for jobs in Europe mostly, but information on the US would also be welcome.

3
  • 1
    As a curiosity, where did you end up with currently? I am currently considering a transition from math to machine learning as well.
    – Paul Pogba
    Mar 25 at 9:00
  • @JamesChung Thank for your comment, but I'll be brief. I did switch first to computational neuroimaging (because of its connection to differential geometry and statistics/statistical machine learning), from where I took an active interest in machine learning as well, and published in the intersection of differential geometry and statistical ML, with application to neuroimaging. Then I worked in industry for a few years, didn't like it and planning to go back to academia, and applying for postdocs/permanent positions in France. Mar 25 at 11:04
  • 1
    Thanks for the reply back. Right now, I am neither familiar with DG nor ML/DL, so it is inappropriate for me to have such discussion at this moment. (I am merely in math master’s) However if anything interesting arises, then I would appreciate it if we can have a discussion (which is likely to be in the distant future).
    – Paul Pogba
    Mar 25 at 13:44
23

1) How hard is this switch going to be? I guess I've all the required mathematics background, but will it be hard to pick up the necessary computer science skills, even if I work in more theory-oriented problems?

Not a problem: picking up the math associated with ML. You've got the right background and you'll find it easy to understand the papers after an initial learning phase.

Potentially a problem: understanding why certain questions get asked and what's considered interesting. This is where mathematicians and computer scientists tend to diverge, and translating your intuition for questions might take some time. But a more mathematical mindset can also lead you to ask interesting questions that CS folks are NOT asking !

What exactly are the programming knowledge I need to master to work in ML?

Depending on how theoretical the postdoc is, anywhere from none to R, python and matlab, and maybe even some distributed large-scale learning framework like GraphLab. But you should definitely get some familarity with the first three - ML is a good example of "no problem formulation surviving first contact with the data".

2) Is there a website/email-list where I can get notifications on jobs in machine learning? I'm looking for jobs in Europe mostly, but information on the US would also be welcome.

One good mailing list is ml-worldwide. Another is the Google group ml-news.

3
  • 4
    To add to Suresh's answer, some people are born programmers, and can pick it up with very little difficulty. Some people (including at least one great mathematician I have met) are utterly incapable of thinking algorithmically, and struggle with the simplest programming tasks. Most of us are somewhere in between. So the answer to the question "How hard is it to pick up the necessary computer skills" really depends somewhat on your natural ability. Jan 26 '14 at 20:25
  • "This is where mathematicians and computer scientists tend to diverge" and both tend diverge from what the scientist who works about the application at hand is intererested in... Feb 6 '14 at 21:40
  • 1
    Another skill that is necessary if you want to get your hands on real problems is translating between the different languages like biologist <-> CS <-> math and so on Feb 6 '14 at 21:41
5

If you are a pure mathematician with background in geometric analysis there are interesting problems in the sub area of machine learning called "Manifold Learning" which requires quite a lot of Riemannian Geometry and intuition. Machine Learning is a vast area and it is a question of what suits you the best.

1
  • Thanks for your valued comment, Mashbat. I've backgrounds in advanced Riemannian geometry, PDE, and Riemann surfaces, but not exactly geometric analysis, which deals with more analysis than Riemannian geometry. I've switched to computational neuroscience with applications of diff. geo., but I'd be definitely curious to know more about what kind of problems they handle in manifold learning that deal with Riemannian geometry. Mar 5 '15 at 12:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.