7

All of the following takes place in a UK university.

I have a BSc in Physics and an MSc in Computer Science. My thesis was on applying various machine learning/statistical techniques to biological datasets. I wanted to do something similar for my PhD, however my supervisor left the university.

I am now in the first year of my PhD in Computer Science, specifically Computational Biology. My work focuses on comparing different techniques (physical/statistical/machine learning) in single cell simulations. I am finding it hard to incorporate machine learning techniques into my work as there aren't many datasets for the kind of thing my supervisor wants me to do and so the machine learning approach is proving tricky.

I desperately want a job/postdoc in a machine learning/stats environment.

  1. Lots of post docs I know switched field after their PhD e.g. Astro-physics to machine learning, dependable systems to machine learning, Biophysics to compiler design. In my case would anyone in the ML community take me seriously? (I thought my Msc would help me out...)

  2. I have taught myself a fair bit of ML and stats, is there anything else I should do to increase the likelihood of getting an ML/stats postdoc?

  3. Would anyone in a stats department take me seriously as I have no maths degree?

  4. Do people that change career areas have successful careers or is this normally a red flag?

  • 1
    Are you planning to do theoretical CS work in your PhD, or will it only be application of known methods? Also, can you clarify what kind of ML/stats postdoc you want - theoretical ML/stats, applied (in what field?), computational biology? – Bitwise Jun 23 '13 at 16:15
  • @Bitwise I'm pretty sure that it will be mostly applications of existing techniques. However if I discovered new techniques I'd definitely publish them. In terms of an ML/stats postdoc, i'm not picky, but given my existing expertise I think applied would suit the most. I would prefer a non biological application, but beggars can't be choosers! – RNs_Ghost Jun 23 '13 at 17:45
  • 8
    is there anything else I should do to increase the likelihood of getting an ML/stats postdoc? — Yes. Publish. – JeffE Jun 23 '13 at 20:25
  • 1
    One question I would ask is if you're in the first year of the Ph.D, but are thinking so far ahead that you want your eventual career to be about ML, why not switch gears and go into ML right now (or after your project) by switching topics/advisors/data sets, or getting collaboration with an ML prof, etc.? – Irwin Jun 24 '13 at 17:16
  • @Irwin this is exactly what I am trying to do, I am trying to make my current work as ML-centric as possible. My supervisor is a big fan of ML, but the problem is getting hold of the relevant datasets. – RNs_Ghost Jun 24 '13 at 21:12
2

Some thoughts on your questions (please don't take any of this as gospel, I am in the final stages of my PhD and are looking for a Postdoc also).

Your ML MSc would more than likely benefit you in any postdoc application (to what extent would depend on the institution). Something to consider, is it possible to build/include ML principles in your current research?

One major way to get noticed in the fields that you are interested in is to get published in peer-reviewed journals and present at relevant conferences. Speak to academics involved in your field of interest, speak to your supervisor/advisor - perhaps inquire if there would be a chance of collaborative papers/conference presentations.

As for changing career paths, this is increasingly the norm - my own example is a switch from economic geology, through teaching to atmospheric physics. One major thing about this aspect is to focus on the skills that you have developed, particularly in research.

I hope this helps.

  • 1
    Thanks, I have the opportunity to teach some AI & ML stuff this coming academic year so I will make the most of that opportunity. Also, that's quite a switch! – RNs_Ghost Jun 23 '13 at 17:46
  • Teaching AI and ML will definitely help as well! – user7130 Jun 23 '13 at 19:59
7

My answer is based more on experience from computational biology, but I think it is relevant for other fields:

  • Changing fields is very common in academia, especially at the PhD/postdoc transition. In many cases it is actually considered an advantage, since you can import your skills, expertise and a certain thought-process into a field in which many people do not have those skills. For example, many physicists, computer scientists and mathematicians have migrated to biology and have made significant contributions. In fact, there are even postdoctoral fellowships that specifically fund this type of field-change.

  • Regarding your "will they take me seriously" questions: Since you are aiming mostly at applied ML/stats, I don't think you should be too concerned if the ML/stats theoretical community take you seriously. Many theorists tend to look down on applied science - don't worry about it, you can still have a significant impact without advancing any theory. It sounds like in the future you will either belong to the department in which you want to apply the techniques (e.g. a biology department) or will work very closely with people in those departments. In this case, you will usually be considered the ML/stats expert.

  • Having said all that, of course it is your job to become an expert. Teaching yourself the theory is important, but if you are going for applied science, especially applied ML/stats, it would be a big advantage to get actual experience in using them. There is a huge difference between learning about these methods and actually implementing and using them. You will see that during your PhD you can often expand your research in directions you are more interested in. It shouldn't be too difficult to use some ML/stats creatively in some sub-projects (which could later be expanded).

  • thanks. If I were to decide that I wanted to stay in a CS department and work on something more theoretical, then I suppose my publications would have to include coming up with some novel techniques instead of just applying existing ones? – RNs_Ghost Jun 24 '13 at 12:38
  • @RRs_Ghost theoretical work would either be coming up with new methods, improving/expanding current methods or general theoretical work. Of course when I say "coming up with a new method", I mean this in the theoretical sense, i.e. mathematically developing the method and proving certain properties of it (e.g. convergence to a good solution with the increase of the amount of data). – Bitwise Jun 24 '13 at 13:00
0

It depends strongly on what you want to do after pursing a PhD degree. More precisely, if you want to work as a technical staff then yes. It affects your career chances because it doesn't help a company that you are an expert, and therefore they have to pay you more than the average, in a different area and you don't have a proven solid background in the area where they want to be hired.

However, if you decided for working as a manager, sale, marketing or administration (e.g. signing applications) then it doesn't matter in which field do you have your PhD. In some positions, it is required to have a PhD title, no more.

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.