I recently transitioned from an academic position to an industrial position to start 123ai.de. I spend a lot of time thinking about the options and consequences, always wondering if it was the right choice. So, what are the similarities and differences when working in an academic environment vs in an industry position in a field like machine learning? Is there any difference between pursuing an academic career and a career in the industry? Is there a way back, once you find yourself in a “real-world” industry position? Does it matter? Well, here is my take on this.
The traditional academic path
Obviously, for pursuing an academic career you gonna need a masters’ degree. In computer-science related fields, it is very common practice to have a part-time student job in the research group that eventually will provide the topic for your master thesis. Ideally (unfortunately) that will not be a teaching position, so you can focus very much on ongoing research in the lab. Ie. you’ll have more time to help PhD students and postdocs with their research if your schedule is free from time-consuming teaching.
Disclaimer: In my opinion, teaching is very important and enables you to hone essential skills such a public speaking and presentation, broadens and deepens your knowledge, and how to educate and motivate (ie. deal with) people. Unfortunately, the current academic system will focus foremost on your research accomplishments (h-index, number of citations, etc) and not on your teaching skills.
If the fruits of your thesis results in peer-reviewed papers, your chances to become a PhD candidate increases significantly. Use this time to collaborate a lot and go deep on at least one topic. Become independent. Go to conferences. Go to summer schools(!). Apply to internships (Google, MSR, etc) and other high-profile groups (preferably in the top US universities). Organize workshops on high-profile conferences.
After graduation, find good postdoc positions in well-known groups. Collaborate a lot and write many research papers to increase your h-index and number of citations. Lead people and let PhD students do the actual experiments (they are time-consuming). Now is the time to try out some teaching. In the meantime, apply for assistant professorships and grants(!). Some of them require at least a two-year postdoc experience. As an assistant professor your single your goal is to become tenured.
The new age (?)
There have always been some industrial places where top-notch research was performed. However, nowadays, there is a very high concentration of top researchers in very few companies, ie. Google and Facebook. This is also reflected in the number of accepted papers at top conferences like NeurIPS and ICML, where companies like Google have more than double the number of accepted papers (170) than the next entity (MIT 79 at NeurIPS 2019).
This, in turn, could also mean that machine learning has finally escaped the research stage and is ready for real-world, large-scale production problems as suggested by François Chollet. Imagine a situation similar to the spreading of computer programming and programmers back in the 60s(?). At first, programming was at the core a math problem. Today, you’ll find plenty of, probably very good, programmers who have only a basic idea of math and this should be considered a big win for the field.
Maybe not surprising, in my opinion the time for industrial endeavors in machine learning has come and it is okay to put the idea of an academic career aside. In this day and age, the communities overlap a lot and there is plenty of opportunities for collaboration and research while solving interesting problems.
If you find this post interesting, please share it and leave a comment below. You can, of course, also support me by donating or becoming a monthly supporter which will come with some more benefits such as exclusive content. If you have any questions, please contact me or leave a comment below.