I studied physics from Faculty to Doctoral student, and now I am working as a machine learning engineer in a different field than when I was an undergraduate student.

When I was an undergraduate student, I was interested in a wide range of fields, not only physics, and attended conferences on synthetic biology (iGEM) and lectured in other Departments of Physics. On the other hand, I was not an exemplary graduate student because I was not very interested in my research and had a hard time deciding on a research theme. I am grateful to my supervisors and others who patiently guided me.

I intended to continue in academia until about the middle of my doctoral program, but considering the employment environment in academia and my own abilities, I decided to work in the private sector and joined an electronics manufacturer as a new graduate in the field of machine learning. I chose this field because it matched my desire to work with mathematical formulas and codes and the demand in society. However, I began to feel that it was difficult to pursue both business contribution and research (novelty), so I decided that I wanted to work directly on real issues without focusing on research, and I changed jobs, focusing on web-based companies, and joined LINE Corporation.

In my current position, I have been involved in several machine learning projects, including a recommendation ^{system*1 and} estimation of topics of interest to users. In addition to developing machine learning models, I am also involved in the implementation and operation of code to run the models in a production environment. The time from the start of model development and experimentation to the release of a model is generally 3 to 6 months, or a year at the most. As a web-based company that operates its own services, we ^also conduct A/B ^testing*2 prior to release. This style of "quickly verifying small improvements, releasing the product, and gradually making improvements over time" was very much in line with my nature.

Thanks to my experience with mathematical formulas in physics, I think the hurdle to understanding machine learning algorithms was relatively low. I also wrote code for numerical computation in my university research. In this way, it may seem at first glance that physics research and machine learning engineering go hand in hand. However, there are significant differences between them.

For example, unlike my university days when one person wrote disposable code, the code I write in my current job is read by others and used continuously. Therefore, code that is easy for others to read and maintain is required. In addition, as long as you are writing code that will run in a production environment, it goes without saying that it is important to write code that will run stably. If you consider that the final product is not a thesis but code, I think you can understand the difference between the two.

On the other hand, I feel that the meta-skills I acquired in graduate school are still directly useful today. When dealing with new issues and technologies, I can say that my dissertation research is almost exactly the same as the research I did in college. In addition, since remote work has become the mainstream, documentation has become more important, and I think my experience in writing papers and the like has come in handy.

What I have written here is only one example, but I hope it will help you to think about your career, whether you are active in the world of science or have moved on from science.

---

Note 1: A system that recommends products or items to users, such as "Recommended for you" or "Users who purchased this item also purchased these items.

Note 2: A test to statistically determine whether a new model is superior to the current model by showing the output of the new model to only some users and comparing its behavior with that of the remaining users.