How I ended up pursuing a PhD in AI at Semcon
If you had asked Jens Henriksson back in 2015 what he would be doing five years from now, the answer would never have been an Industrial PhD at Semcon. Now he shares his story about how an amazing master thesis led him on the right track.
When I started at Semcon as a systems and control engineer in 2015, my knowledge revolved around system modelling and control theory. I don’t even think I knew what AI referred to. The way I learned about it is rather unusual.
My first project here at Semcon was in a team that was supposed to create an offline reference system of laser scans (LiDAR) for autonomous cars. Our part included taking point clouds and segment clusters of points and then try to classify what kind of object it was. As with many projects, there are always some ideas you want to investigate or test, but you will most likely never have time for it. We decided to take some of the crazy ideas and offer them as master thesis topics at student fairs.
How a master thesis wiped the floor with us
A couple of students used one of our ideas about how to classify segments of points for their master thesis. Their results were groundbreaking. It beat both our own solution and the customers’ existing one by far, which of course was very impressive. It also made the customer very happy.
The only problem was that the research included very advanced toolboxes and parallel computing tools that was unusable in the existing developer environment. This meant that I had to break down every single piece of the thesis algorithm and re-implement it to work in the developer environment.
This was the first time I participated in an AI project. Since then, we have had several projects that applied the same technique on different types of data. In fact, the technique is awesome and can more or less be used in any scenario where you have data that represents your problem.
At Semcon, we see a lot of interest in AI from our customers. This is why we participate both in AI research projects, but also applied for an industrial PhD that was approved by the Wallenbergs WASP programme.
My five year project
Currently I am almost two years into my research. My research topic is trying to understand how deep neural networks behave outside of the scope of the model, as well as see if we can accurately get a measure of how certain the model is. In other words, we try to force the system including AI to be certain of an action before it makes it. This is important for safety critical applications, where wrong actions can cause unfortunate events.
While I am still fully employed by Semcon, my work schedule can seem rather unusual. The PhD position includes university courses, sessions with the WASP graduate school, research and conferences. Thus, my work can vary from exploring undiscovered research angles and studying for exams to presenting research papers at a beach in Greece. Overall, the journey is very inspirational, and I enjoy being in the front seat to see and be a part of new research inventions.
Title: ML industrial PhD student // ML engineer // WASP industrial PhD student
Education: M.Sc Systems Control and Mechantronics (Finished) // PhD in verification of deep neural networks (ongoing)
Worked at Semcon since: 2015
I have most fun at work when I…: break down recently published papers with the ML team.