My “previous life” in astrophysics has given me an amazing start as a ML-focused software engineer, who now works in the development of novel medical devices. Throughout my academic career, I knew that I enjoyed programming, and using computers to solve problems. I had been applying this to physics – first quantum chemistry, and then computational astrophysics in grad school. Developing my data analysis and programming skills from the paradigm of physics definitely gave me a unique mix of skills that have served me very well in my current role.
One of the first, most obvious areas is math – my education in physics imparted familiarity with applying numerous different mathematical (and statistical!) techniques to problems, which form the basis of much of my current work.
In addition to the direct transferable knowledge I gained in my academic career, many of the corollary skills have proven invaluable. In my case, this includes programming – which is the primary focus on my work, but can also include familiarity with high-performance computing systems, linux-based OS, and other tools that I used to create end-to-end data processing pipelines to support my research. Being able to take raw data and design a full system that transforms it into useful results is a core skill for machine learning, data analysis, and even generalist software applications!
Having the freedom to work with new concepts in the grad school environment, as opposed to the more rigid class environment, also meant that I could apply them in situations that weren’t artificial, with the goal of using them to facilitate research. In my case, my supervisor allowed me to pursue my own interests with regards to methodology, which gave me unbridled time to dedicate towards machine learning concepts with a focus on applying them towards our group’s data. This scale of free learning environment is harder to replicate in a company, particularly when your billable time is the main source of revenue.
Finally, several soft(er) skills have also proven to be very useful in my new role. I can find and read academic papers without issue, which is a niche but powerful skill for R&D situations where a novel methodology, or the entire scientific basis for a project is present only in literature. The constant presentations that grad students do are also a crucible that give young scientists the tools to summarize information for wide varieties of audiences. Being an engineer who can also communicate effectively, or give good presentations, helps you stand out. Last but not least, the structure of smaller development teams is a close analog to research groups. Regular “stand-up” meetings, and the ability to rely on your other group members for help with tough problems is a close parallel between the two regimes.
My thesis was focused on the use of deep learning to enable astronomers to use stellar chemistry to identify stars that formed outside of Milky Way-like galaxies. I established a process of data selection and processing, model definition and training procedure, and finally some model validation methods. All of this makes up a major component of my current work, and are skills that were nurtured during my PhD.
Thorold Tronrud, PhD
Software Engineer, StarFish Medical
Notes:
Physics + Astro increased familiarity with mathematical methods
Work on analysis of computational results means learning data processing and analysis via programming
Supervisor allowing me to explore my own interests w/r/t methodology (e.g. using NNs) meant I could dedicate a lot of time towards learning how they work and how to use them
As SW/ML engineer:
Development-based skills:
I’m entirely unafraid to dive in to math, which has been handy many times during my work
The programming I did during my grad school work to support my research maps very well to ML and data science applications – being able to design and create a full pipeline for any type of data is very useful
Exposure to ML concepts in a much less limited environment than classes meant I also got practice applying the techniques in situations much closer to “real life”
Softer-skills based:
I can find/read academic papers – a highly underrated skill, but incredibly useful in R&D situations where methodology, scientific basis for the project is present only in literature
Presentations – being an engineer who can give good presentations and summarize information well for a wide variety of audiences helps you stand out
Research group structure applies quite well to development team structure – regular “stand-up meetings”, asking team members for help with tough problems, it’s all parallel with aspects of non-academic R&D