Jinya Sakurai leveraged his exchange study and internship experiences to further his career and interest in machine learning. Find out how.
Humble beginnings
Unlike many students at the University of Tokyo, I attended a public school rather than a private school. Students in our school had diverse economic backgrounds, and their futures varied. Some of us went on to study at universities, while others could not for several reasons.
My high school teachers encouraged us to apply to universities entirely based on their reputation. I studied hard to get into famous universities in Japan. After passing the entrance exam, I joined the University of Tokyo, made friends, and was impressed by how computer science and machine learning help modern society. That is why I picked the Department of Informatics to study computer science, programming, and machine learning.
Computer graphics internship
Two years ago, I applied for an internship at Albert Inc. Japan, a company that develops computer graphics and machine learning algorithms. I did not know much about computer graphics then, but I was intrigued. They gave me a recruitment test related to machine learning and deep neural networks. And based on my performance, they accepted me. Since it was during the COVID pandemic, the internship was online. My mentor helped me understand how to use neural networks to represent 3D shapes.
During the internship, I learned that if you are stuck in your research, reading more papers might help because somebody else must have asked those questions.
International exchange study
Every year around September, the University of Tokyo emails us about the University-wide Student Exchange Program (USTEP*1). At that time, I was so busy with my homework and assignment that I ignored the email. Then there was a second email reminding us to apply. I had been wanting to do an exchange study, but it would mean I needed to extend my course here by one year. It was a difficult decision to make. With the encouragement of my parents and supervisor, I applied. I had to prepare a CV and a motivation letter and take an English language test, but I succeeded.
The Japanese government awarded me a stipend to study at the Royal Institute of Technology in Sweden for ten months. At the institute, I studied Swedish, topological data analysis, and probability analysis, conducted research, and improved my English. I took mathematics courses because I am fascinated with mathematics as a tool to analyze and formulate data science and machine learning problems.
Life in Sweden
The winter in Sweden was so severe and depressing as the sun set around 3 pm. But I took vitamin pills and played badminton, which helped. I felt like I was more robust than I expected; I could adapt to a completely different environment.
Swedish people are, like Japanese people, humble and polite, but the biggest difference between Japan and Sweden is the party culture. They had places on campus where students could dance and party. It was a pleasure to experience a completely different culture of the university. Another surprise was that most people could speak English even though that is not their native language.
In Sweden, master's students can work with the industry for their thesis and solve practical problems. I was paired with a master’s student who was working on cloud segmentation. Given a cloud image, they wanted to predict the amount of cloud coverage in the image and help forecast weather. It was an invaluable research experience that stoked my interest in applying mathematics to machine learning.
Continual learning
After the exchange study, I returned to Japan and worked on my senior thesis about a machine learning model called continual learning. Machine learning models like ordinary image classifiers are trained on fixed data sets. But, over time, due to security or memory restrictions, we must delete data and free up memory, which changes datasets and tasks. My research aims to build a machine learning model that can adapt to new data while not forgetting what it has learned.
Consider an algorithm trained on a textbook. In a true continual learning scenario, even after being trained on this textbook, the algorithm would still have the capability to learn from new sources without forgetting what it has already learned from the textbook. We achieve that by pruning the network, like carefully trimming a tree.
I presented my thesis in early August. Before that, I did a practice presentation in my laboratory. My professor was strict and highlighted many errors in my presentation. I was scared. I could not sleep before the final presentation, but it went well.
The way forward
I could earn all the credits required to graduate and finish my thesis by the deadline, which helped shorten my extension by six months. I wanted to use the next six months meaningfully. So I applied for an internship in Saudi Arabia. But I was surprised when they asked for three recommendation letters. Fortunately, I could leverage my connections from the exchange study and internship to get the letters. I am now ready to experience the new culture and climate in Saudi Arabia as I travel there in September 2023. The experience will help me decide what to do for my master’s degree and how to forge ahead in my career.
Ph.D. or not?
I talked to my friends and lab alumni about whether I should pursue a doctorate. I understand that Ph.D. students do not earn much, but one can get into academia with a doctorate. Working in the industry after a master’s degree often pays more, but you will not get a doctorate even though you might do research. When deciding whether to pursue a Ph.D., you must be prudent and consider many things. I will cross the bridge when I get there.
▼USTEP*1
https://www.u-tokyo.ac.jp/en/academics/ustep.html
※Year of Interview:2023
Interview and text: Ravindra Palavalli Nettimi
Photography: Junichi Kaizuka
The interview was edited for brevity and clarity.