Disclaimer: machine translated by DeepL which may contain errors.

After completing my Graduate School of Science's Department of Physics, I joined NTT Laboratories. I then transferred to NTT Research, Inc., NTT's U.S. base, and am currently working in a laboratory at Harvard University as a Research Scientist.
During my master's program, I belonged to a laboratory for particle experiments and worked on data analysis at CERN. However, my desire to "pursue my own research theme" and my goal to "earn money in the future" led me to apply for a research position in the field of machine learning. I was attracted to NTT Laboratories because of its highly flexible research environment. After joining the company, I worked on technology development and research in the fields of machine learning and data mining. I was particularly attracted to research, and after my fourth year with the company, I shifted my focus to research-oriented work. After five years with the company, my passion for science did not disappear, and I became interested in the boundary area between machine learning and science, and began to write a dissertation on the theme of "AI for Science. At this point, I entered a doctoral program at Kyoto University and earned my Ph. People tend to think that a doctoral program while working is difficult, but since I earned my degree through a Doctoral dissertation I wrote while working, the burden was not as great as I had imagined.
Traveling to the Grand Canyon
I had always had a strong desire to do research in the United States, the home of this field. Therefore, after obtaining my doctoral degree, I asked my supervisor if I could transfer to NTT's AI research base in the U.S. He readily agreed, and in 2023 I was transferred to NTT's AI research base in the U.S. He readily agreed, and I transferred to NTT Research in the U.S. in 2023. Currently, I am working on research as part of the CBS-NTT Physics of Intelligence Program, a joint research program with Harvard University. The theme in Japan was "AI for Science", but in the U.S. it has changed to "Science of AI". Currently, I am working on the theme of scientific understanding of the behavior of AI models (language models and image generation models). In particular, I am focusing on the phenomenon of "emergence," in which performance improves dramatically as the model size is increased. This phenomenon is a key theme for AI applications, but the timing and causes of the performance improvement have not yet been clarified. To address this issue, we design simple artificial tasks that capture the essence of the phenomenon and explore its performance and underlying mechanisms through training and prompting using small-scale AI models.
We believe that scientific understanding of AI models is very important in today's world, where we aim to control AI and coexist with society, and that it is of great social significance. The design of simple artificial tasks is a physics-based approach, which draws on my experience in experimental physics that I worked on as a student. In addition, I enjoy conducting experiments on AI models as unknown intelligent life forms because it is a very interesting scientific pursuit and fits my personal interests.
I feel that curiosity has been the driving force in choosing my career path, and it has led me to my current fulfilling job. I hope this experience will be helpful to you.