February 27, 2024

Appeal of creating technologies used on the front lines of society

Senior Researcher, Visual Intelligence Research Laboratories
Yasunori Babazaki

After completing a doctorate program in astrophysics (X-ray astronomy), Babazaki joined NEC in April 2018. After being assigned, Babazaki was involved in researching and developing image and video recognition technologies using deep learning and has produced many technologies with a high degree of utility and applicability such as the development of a solution for automating the length measurement of farmed fish and a technology which combines information related to behaviors on a construction site in a multi-modal manner to understand context.

Joining NEC with the goal of creating technologies "that someone will use"

The reason why I became interested in working at NEC is the NEC internship that I participated in during my doctoral program. At the time, I was researching the observation of galaxy clusters and supernova remnants, so I thought that if I learned NEC's image analysis technology, I might be able to apply it to my own research and decided to participate. What made an impression on me during my internship experience was the positive atmosphere inside the company. The people around me answered any questions that I asked, and they kindly provided consultation. I remember feeling strongly that NEC had really developed a climate for conducting research.
For that very reason, it was natural for me to choose to join NEC after finishing my PhD. Of course, I did consider continuing research in academia, but when I probed my own motivation for research, I realized that I was deeply attracted to the idea of "having someone use" the results of my research. That being the case, I thought that becoming a corporate researcher would be exactly the right position.
In addition, depending on the topic, X-ray astronomy is a field with a long research span. It is the kind of world in which you can spend more than 10 years including the development of the astronomical satellite needed to obtain the observational data before one research study is completed. In contrast, much of NEC's research reaches commercialization in a span of about three years. The ability to produce many research results that are used in society was one deciding factor.
Plus, it is not the case that I stopped being involved in academia just because I joined a company. In fact, NEC is a company that has had many research papers accepted by top international conferences. I chose NEC because I was drawn to their culture which encourages both business and scholarship.

Statistical knowledge applied to physics proves useful in machine learning

I was coming from astrophysics and taking on the challenge of image and video recognition, so in the beginning I was a little worried about whether I would be able to catch up. However, when I actually started to research machine learning, there were many areas in common with my previous research, and I was able to easily get into it faster than expected. Specifically, I was able to directly apply my knowledge of statistical analysis and data fitting from astrophysics to machine learning. I hope that people who have majored in astrophysics and want to take on the challenge of the AI field will become interested in NEC.
Later, my senior colleagues taught me various things and with their help I was able to win the Gold Prize of the 2022 Field Innovation Award of the Japanese Society for Artificial Intelligence for developing an automatic farmed fish measuring system using image recognition AI. This system measures the size of farmed fish during migration from video using AI. In addition to becoming able to increase the frequency of fish size measurement during cultivation and precisely understand the growth process, it also reduced the burden on the people measuring the fish.
Naturally, I was happy that the image recognition AI that I was in charge of was well received, but I was extremely happy that a system that was actually being used by customers out in the field was recognized for its value. It was the moment when "having someone use" a research result came to fruition.
That was also the case with this technology, and the difference between NEC and academia is that it conducts research while collaborating with business units and customers. The process of digging up customer needs and issues while engaging in mutual discussion to create technologies that can actually be used is unique to corporate research and extremely interesting.

An environment in which researchers have significant discretion

As a place that researches image and video recognition, I think that NEC is an extremely good environment. That is because it has a complete set of assets for machine learning. For example, NEC owns the largest-scale AI supercomputer in Japan with a specification of over 580 PFLOPS, and it is able to obtain a lot of the data which is necessary for training AI systems. Because NEC is developing businesses in diverse business areas, it is able to utilize live data that is directly linked to social issues.
In addition, as I mentioned a moment ago, I felt during my internship that NEC had developed a research culture, and after I actually joined the company, I keenly felt that it was an environment that really respected the opinions of the researchers. As a specific example, there is a system called the "20% Rule*," which allows researchers to devote up to 20% of their work time to research a theme of their choice. Being able to independently pursue research with a slightly longer-term perspective that is beneficial for the future, using this rule, is extremely attractive.
In addition, there is also the "Zero-order Proposal" system. This is a contest in which researchers present a new research theme of their own conception, and if it gets approved, a research budget is assigned to that theme. Allowing researchers to pursue not only the themes determined by the company but also respecting the independence of the researchers by granting significant discretion is one of the great things about NEC.

A day at work

Message to my past self in my school days