Two things I had to unlearn on the way from astrophysics into industrial machine learning.
When you finish a PhD or a postdoc and someone asks what comes next, there are usually two answers available. Stay in academia. Or go to industry.
In STEM, that second option collapses further, almost always into data scientist or machine learning engineer. That is the entire map most of us are handed.
I remember what that map did to me. If you have spent years following your own curiosity, the industry side reads like a warning. A bank. A dashboard. A set of tasks somebody else has already defined and now needs executed. The choice appears to be curiosity on one side and a salary on the other, and if you are the kind of person who values the curiosity, you stay in a system that may not have a permanent place for you, because the alternative on offer looks like giving up the thing that made you do this in the first place.
It took me a long time to notice I was answering the wrong question. The question was never academia or industry. It was this: do I want to keep doing research, and if I do, where does research actually happen?
Unlearning one: research is a method, not an address.
Research happens in far more places than the university, and almost nobody lists them for you.
There are non-university public institutes. There are applied research institutes, which sit deliberately between fundamental work and something a company can use. There is corporate R&D. There are deep tech startups, and contract research organisations that do nothing but research for other people. Different funders, different timescales, different definitions of what counts as finished. The same method underneath all of them.
I did my PhD in computational astrophysics at the University of Cologne, modelling how the diffusion of molecules shapes star formation. I now lead the development of a physics-informed machine learning system that predicts where defects will form in cast metal components, before anyone pours any metal. Different address entirely. The same work: form a hypothesis, design the thing that could disprove it, argue honestly with the data, then find out you were partly wrong and go again.
I was a researcher then. I am a researcher now. The name of the university was never the thing making it research, and it took me an embarrassingly long time to accept that. Once I did, the decision stopped being an identity crisis and became something much smaller and much more answerable: which address, on whose funding, for how long.
Unlearning two: papers are not the measure of impact.
In academia I always knew what a good year looked like. Papers published. Sometimes other things, but mostly papers.
Outside, that metric does not disappear entirely. In applied institutes and in some corporate research groups, publications still count for something. But they are no longer the measure, and if you keep using them as your scoreboard you will feel like you are failing while doing perfectly good work.
The measure is what the research turns into. Somebody needs something. You build it, and you show them what you built. The pure advancement of knowledge is not the goal in itself here. The knowledge gets taken out of its original context, applied to a different one, and transformed into something a person can actually use in their day.
That transformation is the part I did not expect to care about as much as I do.
My work now sits between a solver and an engineer. When someone looks at a prediction my system produced and changes a design because of it, before a single part is cast, that is the result. Not the accuracy figure. Not the write-up. The decision somebody made differently because the work existed.
There is a purpose in that which I had not anticipated and did not have before. In astrophysics I was adding to a body of knowledge, and I loved it, and the timescale on which it reached anyone was measured in decades if at all. Now the distance between the work and its use is short enough that I can see it. That closeness is what makes the second unlearning worth it, and it is the honest reason I have not looked back.
So, which one is it?
I think most people stuck at this decision are stuck because they have merged two questions into one, and the merged version has no good answer.
Are you leaving academia, or are you leaving research?
If it is academia, then what you are looking for is an address, and there are more of them than anyone told you. The methods you spent years acquiring travel with you.
If the honest answer is that you are leaving research, that is a legitimate answer and not a failure. Some people discover that what they loved was a specific question, or a specific group of people, or teaching, and that the method itself was never the attachment. Knowing that saves you from taking a research job somewhere else and being disappointed in the same way for different reasons.
Either way, answer it for yourself before you answer it in an interview. You will be asked why are you leaving academia by someone across a table eventually, and I have written separately about how to answer that question without sounding like you are running from something. But that answer is only convincing if you have already worked out which of the two you are actually doing.
I was not leaving research. I just needed a different address, and a different definition of what the research was for.
Dr. Aleena Baby will be speaking at our upcoming webinar on September 17, Researcher Impact: From Academia to Industry. If her perspective resonated with you, join us live to hear more and ask your questions directly. Register now for free.
Dr. Aleena Baby is a computational physicist and machine learning engineer working on physics-informed ML for industrial applications at ACCESS e.V. in Aachen. She writes at Academia to Industry and coaches STEM PhDs moving into industry roles across Germany and the DACH region.
