3 routes to combine machine learning with domain-specific experience
One of the most common questions I get is from medics interested in machine learning, wondering how to get started.
I’m currently in the process of writing a longer article about getting started with machine learning.
But today I want to consider a more specific sub-question: Do I need a Master’s degree to get involved in Machine Learning (ML)? If not, what are the alternatives?
👨🎓 Do I need a master’s degree in data science or machine learning (ML)?
In short, it depends what you want to do with machine learning.
I’m going to consider three of the routes that I’ve considered and try and extract some general points. Those routes are:
Applying ML to healthcare in academia
Starting a digital health start-up
In a full-time technical role (such as a data scientist)
I want to preface these considerations with the caveat that I’m not an expert on this, and these are just some of my personal thoughts. I may not agree with myself a year from now. But as of right now, my thoughts are as follows:
🏫 (1) Applying ML to academia (e.g. in healthcare)
In academia, the best approach is to find a problem, and then identify the solution.
Sometimes the solution will involve ML and sometimes it won’t.
In the spirit of being problem-led, the benefit of understanding ML depends on the problem you are trying to solve.
Typically, when addressing a clinical research problem with AI, you only need to understand the ML that is specifically helpful to that problem.
Let’s say you are a clinical researcher at Moorfield’s Eye Hospital, looking at the use of convolutional neural networks to recognise retinal disease. It’s really helpful to understand the approaches to Computer Vision, but less so approaches to NLP, probabilistic graphical models or bioinformatics, for example.
If you are serious about academia, taking a PhD will provide a problem-focussed way to understand the ML relevant to your problem.
I believe a master’s will give you a much broader foundation of technical understanding, but this may not be necessary. Sometimes it’s a useful bridge to a PhD, but it’s also an extra year of your life.
MY CONCLUSION: Probably not necessary. Perhaps a useful bridge to a PhD.
👩💻(2) Starting a start-up (e.g. in digital health)
This is a trickier one, as I don’t think there’s one “route” to launching and running an impactful start-up.
In my head, the two key things are:
Good insight into the problem and how it can be solved
Good execution of the solution
For the first part, working in clinical medicine is invaluable for identifying problems for healthcare. Likewise for other industries. Understanding different forms of technology helps you to understand potential solutions.
So, like with research, it’s a case of using the technology most appropriate to the problem. Sometimes this will involve ML, at other times it won’t.
Another important consideration is that start-ups are composed of teams of people with as diverse skills as possible. If you have domain-specific insight (such as in clinical medicine), then that’s the main value you bring to the team. You’re not going to be the best clinician and the best with machine learning, so there’s going to be someone else on the team with that skillset.
So your role therefore becomes more about communicating what needs to be done, rather than actually creating the technical solution. You need to be able to speak the language but not necessarily much more beyond.
If you know what problem you want to solve, ask what technology is best placed to solve it. Then go learn that.
If you are less sure on what you want to do, going deep into ML can be a good way to hedge your bets. It may provide more insight, and help you identify deeper technical solutions to existing problems. This can end up as a long-winded way to do so, however, so it depends on the individual.
MY CONCLUSION: Not essential but may help you identify new ways of solving existing problems (particularly if they have a complex technical solution). Depends on problem you end up trying to solve.
📱(3) Full-time technical role
You may be open to working full-time in a technical role, such as a data scientist or machine learning engineer.
There are a lot of data science roles that don’t specify a master’s degree as a formal requirement, and I think it’s possible to get a job without one.
However, for me, and coming from an unconventional background, I found it hard to get taken seriously until I started working towards one.
I think if my bachelor’s degree was more directly relevant to data science (rather than Medicine), I may not have needed to. A good bachelor’s + proof of practical experience is sufficient in many cases.
I ended up choosing a master’s degree in Data Science and Machine Learning at UCL in London. Even before starting it, I was pretty confident that I had a good grasp of key concepts from my own self-study.
But a lot of job applications led to straight-out rejection, so I never had the chance to prove myself. And I can completely understand why. If you have a lot of applications for a role, the one that says “I’m a full-time doctor, but have done loads of data science self-study” is a pretty easy one to remove from the pile.
The master’s didn’t guarantee I’d progress to an interview, but I definitely felt it helped me get a foot in the door.
MY CONCLUSION: Probably worth it if from non-conventional background and serious about working in technical role. Doing a bootcamp is an alternative (as per a previous email) but weaker support for the transition.
📝 In summary
For academia, if you know what area you want to work in, master’s may not be necessary
For start-up, if you know what problem you want to solve, consider whether it needs ML + consider your role in the start-up. If yes, could be worthwhile. If don’t know what problem, master’s may help you identify it.
For full-time technical role, and if you don’t have a clear background, it could be the gateway that you need.
These are just my two cents. As I said, I’m going to be thinking more about this and fleshing out these ideas more formally over coming weeks.
Comments powered by Disqus.