A Sense of Systems and the Power of Teamwork | Yusuke Doi (Class of 1998, Faculty of Environment and Information Studies; 2000, Master's Program, Graduate School of Media and Governance)
2021.08.27
Yusuke Doi
Executive Officer, VP of Computing Infrastructure, Preferred Networks, Inc.
/ Ph.D. in Information Science and Technology (The University of Tokyo)
Graduated from the Faculty of Environment and Information Studies in 1998
Completed the Master's Program at the Graduate School of Media and Governance in 2000
There is a list called the Green500 . It ranks the world's Top500 supercomputers by power efficiency (the number of calculations per watt). The Top500/Green500 lists are published twice a year, and the MN-3 computer, built by my company Preferred Networks (PFN), ranked first in the world on this list in June 2020, second in November of the same year, and first again in June 2021.
Power efficiency is fundamentally crucial for computers. A computer's power efficiency is almost directly linked to the limits of its performance. Therefore, for computers built with the same level of semiconductor technology, a more power-efficient design results in higher performance. In fact, the MN-Core accelerator (a dedicated computer for accelerating large-scale computations), which is used in the MN-3 and was also designed in-house at PFN, has achieved higher power efficiency than other companies' accelerators built with more advanced semiconductor technology. This has shown that design technology alone can have an impact significant enough to overcome a generation of semiconductor technology.
I believe that for various industries, the importance of computers and computational power will continue to grow. Machine learning, including deep learning, is making it possible to tackle the real-world challenges faced by industries. An era is beginning in which we will harness the power of computers in a wide range of fields—not just image recognition and language understanding, but also chemistry, drug discovery, manufacturing, and logistics—by securing an unprecedented amount of "computational volume" to model and solve our problems. Under these circumstances, we believe that computational power will create a decisive competitive advantage in industry, and we are trying to build both the computers and the industries that will use them.
As I wrote at the beginning, the MN-3, a computer we built with this philosophy in mind, has become the world's number one computer by a certain measure. And I am now responsible for PFN's entire computing infrastructure, including this world-class computer. While it's true that at SFC I studied the internet in the Murai Lab and conducted research and practice that included distributed systems, I originally had no connection to supercomputers. At the time, PFN was still a small company, and many of its systems were still under construction. In that environment, I ended up applying the "system administration" skills I had developed as a student, taking care of the information systems and computing infrastructure, and my current career has been an extension of that.
Looking back, I feel that the word "system" has been the key to understanding my journey from SFC to the present day. Of course, "system administration" is not the same as the word "system" in its general sense, but I believe that whether or not one has acquired a systemic way of thinking makes a big difference when building things in the real world.
During my time at SFC, I was in the Murai Lab researching a technology called the Domain Name System. As an undergraduate, my perspective was still narrow, and I just holed up in the lab, digging deep into a specific technology. However, I recall a moment during the first year of my master's program when my perspective suddenly broadened. This happened in a class (I believe it was taught by Professor Iseki, but my memory is hazy as it was over 20 years ago) where I encountered Kuhn's theory of paradigms and became engrossed in the works of von Bertalanffy, Wiener, and others. Of course, it wasn't just the classes; I think it was also beneficial that there were people around me working on completely different things, allowing me to get a broad overview of many fields just by being on campus.
Later, as I pursued a career as a technology researcher, I came to realize the value of the "systems" way of thinking, which is built on the combination and relationships of different components, people, and values. It seems that even outstanding engineers and researchers, or those with exceptional, internationally recognized abilities in areas like mathematics and algorithms, often cannot see the whole picture on their own when it comes to integrating components to build something as a system. Perhaps my SFC-style training forms the foundation for how I approach integrating different perspectives and viewpoints. On the other hand, it is also true that there are profound worlds that one cannot see or reach if one remains confined to SFC alone.
To put it bluntly, no matter how much you study at SFC, you won't learn how to build a computer with the world's best power performance. However, if you can acknowledge what others do better, avoid setting boundaries for yourself, greedily absorb what you can, and contribute your own strengths for the sake of an excellent team, you can become a member of a world-class team. And perhaps, you might even be able to change the world using a world-class computer. SFC is physically isolated in the hills of Fujisawa, but now that the world has rapidly gone remote due to the coronavirus, this might actually be a chance to step out into the wider world.