0

I am a research scholar commencing my research in the field of deep learning. Deep learning uses lot of tools from mathematics. To name a few : linear algebra, probability and statistics, optimization.

My mathematical background is very shaky. When I read research papers in deep learning, I invariably skip the sections that are mathematically heavy. However, I can understand the higher level details, such as what the research work is about, what algorithm the authors have used, and so on. But I never have the feeling of completely understanding the paper.

I feel that I cannot do original research unless I become really strong in the above mentioned subjects in mathematics. Is this claim correct?

That said, can someone publish papers in deep learning without really understanding the mathematics behind it?

  • 1
    You're starting your research career. So I don't think you should expect to fully understand all the details of papers. You'll pick-up enough of the skills you need along the way. – user2768 Jan 30 '17 at 13:12
  • I suspect you probably can do original research in deep learning without being really strong in those subjects, but learning those subjects better will undoubtedly help with your research. Try to learn them as well as you can. – Peter Shor Jan 31 '17 at 2:51
  • Look up some courses online and see the level of mathematical maturity expected of students in this area of research For example, this is an assigment for an undergraduate introduction to machine learning at University of Toronto which has became some sort of a hot spot for ML research psi.toronto.edu/~jimmy/ece521/a1.pdf How well do you think you would do on these assignments with your current level of mathematical maturity? Then fill in the gaps. – Bajie Feb 26 '17 at 21:16
2

I feel that I cannot do original research unless I become really strong in the above mentioned subjects in mathematics. Is this claim correct?

If it is, then what would you do about it? Probably, take one math class per semester, and in the summer too if possible, and do some additional reading on your own when classes are not in session.

If the claim is not correct... what harm would it do to strengthen your math skills, even if they turn out not to be essential for your field?

| improve this answer | |
  • Valid point. I guess I should start working on my math background right away. Thanks! – Karthik Thiagarajan Jan 31 '17 at 3:32
  • Linear algebra is a great place to start, if you can find the right course offered when you need it. You'll get to use logic and become more comfortable writing proofs. (Just make sure you choose a course that does require proofs -- don't take a cookbook version designed for engineering majors.... – aparente001 Jan 31 '17 at 3:36
  • ...Probability will be fun. That would be a good one for the summer. // If you've already done the calculus sequence and an introduction to ordinary differential equations, then at some point it might be helpful to take advanced calculus -- mainly because it just gets you to a slightly higher level of sophistication, not because you would necessarily need exactly the topics that are in that class. Don't hesitate to show your transcript to a friendly applied math professor from time to time, to check that you're on track. – aparente001 Jan 31 '17 at 3:37

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.