March 18, 2020

In 2020, the academic research paper search service “AMiner” developed by China’s Tsinghua University announced the “AI 2000 Most Influential Scholars (AI 2000),” and NEC’s Hiroaki Inoue was included on the list. Since joining NEC in 1999, Inoue has been working as a researcher in the field of computing. He has continued to achieve remarkable success, such as when he was selected as a Best Paper Award Nominee, of which there are still few in Japan, in 2006 at the prestigious ACM/IEEE Design Automation Conference (DAC). What kind of experience and mindset produced these research results? We heard the details from Inoue himself.

Valuable research that straddles the boundary between hardware and software

― How does it feel to have been included on Tsinghua University’s “AI 2000 Most Influential Scholars (AI 2000)” list?

Honestly, I was more surprised than anything (laughs). Needless to say, I am extremely pleased that the research results that I have been working to achieve with various collaborators have been praised in such a fashion. Still, I never imagined that I would receive this kind of attention, so I was very surprised.

― What do you consider to be the main reason you were included on the list?

I’ve heard that this list was derived by Tsinghua University’s own algorithm based on the number of research paper search hits, quotations, and other data. As a result, I cannot really speculate as to what the main factor was, but I have a feeling that the research paper that I worked on with an American start-up several years ago had a significant impact. That’s because the names of the researchers who worked with me on that research have been included on the list as well.
The research paper dealt with platform technology for improving the efficiency and speed of AI processing, and it was accepted at the USENIX Symposium on Operating Systems Design and Implementation (OSDI), one of the top academic conferences for system software. This research paper has been quoted more than any other paper I have written—over 1,600 times already. This is a full digit more than the average number for my research papers.
However, not all of the authors of this paper were included on this list, so I believe that my other research results have been evaluated favorably as well.

― What types of research have you been involved in thus far?

I have been engaged in research related to computing. Specifically, my research concerns semiconductor chips and the related system software. Because this field exists on the boundary between hardware and software, there are actually not that many people who can perform the research. Thankfully, I have been able to deal with both areas throughout my research, so before I knew it, I realized that this was where I could take advantage of my strengths and decided to make it my focus. As for how this relates to the AI technology that was selected on this occasion, the main research theme concerns the part of the platform that provides support for efficient operation of AI programs.
My research at NEC began with research on small chips for the mobile phones that were rapidly becoming widespread at the time. How does one decrease power consumption to make the battery last a long time while also ensuring speedy operation? This is the kind of research I was engaged in.
Also, that period was one where there was a global focus on the issue of how to raise the reliability of the chips themselves. To give the simplest possible example, we had to figure out to what extent we could maintain normal computations without using malfunctioning chips. And upon solving this issue, how could we create durable chips that would not break down? This is the technology development that the world was focused on. But actually, at that time, I was confident that the software technology that I had developed could make up for chip malfunctions. However, there was no technology for detecting the all-important malfunctioning chips. There was, however, a professor at Stanford University in the U.S. who possessed extremely effective technology for detecting chip malfunctions. I thought that using their technology to detect chip malfunctions and using my software to make up for those malfunctions could resolve the issue. Based on that plan, in 2007 I used the company’s study abroad system* to spend a year at Stanford University as a visiting researcher. It was a truly stimulating year, and I learned so much. This study abroad experience significantly changed my approach to research, and it may have had a considerable impact on my being selected for this list.

“So what are you going to do tomorrow?” The impact of Stanford University

― What kind of research did you do at Stanford University?

First off, the research was shocking from the start. When I first met the professor, I handed him a research plan that I’d created for the year. But he didn’t indicate any interest in it at all. In the end, he began folding it up as we were talking. When I wondered why, he asked me, “So what are you going to do tomorrow?” This was really shocking to me. After all, I had gotten used to doing research at a company, so I thought that pursuing my research while sticking to the mid- to long-term schedule and milestones that had been set was the norm. I had never been asked or even considered what I would be doing the next day, without even creating that kind of plan at all.
In reality, the research I did afterwards diverged significantly from my original plan of action. I held discussions with the professor once every few days, with him suggesting that I do something a different way, me agreeing, and then him telling me to go do it. The speed with which we acted was completely different. It was the agile style of development that has become a topic nowadays, and we pursued our research while maintaining close communication and frequently adjusting our direction. This kind of speed came as a big shock.
Another shock was the extraordinary attention given to the way research papers were written. Stanford University’s motto is that those who do good research should write good research papers, and there was this deep obsession when it came to papers. I actually co-authored a paper with the professor while I was studying abroad. I began my preparations about a month in advance and wrote a draft, and I figured it would be done after about 10 revisions or so. It ended up going through about 100 drafts. Besides the obvious points, such as making sure the logic was sound and the results were conveyed correctly, it was checked step by step multiple times, down to every word and phrase, to ensure that even those who saw it for the first time would be able to clearly understand the significance and value of the results we obtained. In addition, the professor went as far as to ensure that I read the research papers by the peer-reviewing members of the conferences we were submitting to for publication so that the paper could reflect the latest technology trends. I remember being extremely surprised at this level of attention. To be honest, before studying at Stanford University, I thought that the name of the university alone would be enough to get my research papers approved without issue (laughs). In the end, the research paper I co-authored with the professor, “VAST: Virtualization-Assisted Concurrent Autonomous Self-Test” was approved by the prestigious IEEE International Test Conference (ITC). The professor told me that the paper was received quite well at the conference. Incidentally, the content deals with technology for raising the reliability of chips, which was the very reason I wanted to study abroad in the first place, and it formed the base for my future research.
I was also surprised by the expansiveness of the U.S. researcher community. Once, I was participating in a conference and mentioned to the professor that a research paper by someone from a certain company was quite good. The following week, the primary author of that research paper showed up at our lab. I was invited there to take part in a discussion. When I asked if the professor knew this person, he told me that he did not. This was during the period before social media had fully developed, yet apparently there was some kind of community at the time that allowed for researchers to casually reach out to one another. In fact, a variety of people frequently visited the professor and kept in touch with him. But in order to be a part of this community, you had to demonstrate what you could contribute. There were times when I was invited, and there were times when I was not invited to certain gatherings. Thinking about it now, I think the community at that time was similar to the open source community that exists nowadays. I learned a lot about building these kinds of human networks.

Pursue your research while trusting your instincts

― How do you go about producing results through your research?

I think there are a lot of factors, but I believe first and foremost that you should trust your instincts as you pursue your research. In 2006, I was chosen as a Best Paper Award Nominee at the leading ACM/IEEE Design Automation Conference for a research paper I had produced before studying abroad. However, these results were produced not by research assigned to me by my boss, but rather research I had pursued on my own. NEC has something called the “20% Rule,” which is a system that allows you to pursue your own research as long as it occupies 20% or less of your work hours. I used this system to do my research.
I worked on the theme for this research with an eye toward the future while holding discussions with related divisions both inside and outside the company. With mobile phones at the time, applications offering various functions besides phone service became available at last, but only the applications provided by mobile carriers were available for use. Yet I suspected that in the future, we would have a world where users could freely use a variety of applications just as with computers, including those from third parties. From today’s perspective, this seems like a matter of course.
What poses a risk in such a world are viruses and other malicious programs. So I researched and put together a paper on the technology that would respond to this risk by using software to create a virtual environment that isolates dangerous applications along with hardware that performs thorough operation checks in order to ensure safety. This technology is used nowadays as well, and I believe it was an idea that took advantage of my ability to understand both hardware and software.
Actually, this technology is related to the aforementioned results of the collaborative research I conducted with Stanford University. In a sense, the idea that software malfunctions lead to viruses developed into the idea that hardware malfunctions lead to defects. This research made me realize how one begins with a single powerful technology, then links that with another type of technology, with these individual points eventually forming a line, and these lines subsequently linked to form a surface. This progression is how research develops and how the fruits of that research gradually expand.
That being said, this research did not go as planned in the initial stages. My boss at the time put pressure on me, asking me if what I was doing had any significance. I used these views to motivate myself and continue my research, which resulted in me being able to produce good results. I think trusting your instincts and pushing forward is an important approach.
In addition, I always asked myself if I was aiming for the best and most unique technology. I analyzed my thinking and told myself that if this was not my goal, then there was no point to what I was doing, and thus continued pushing my research forward.

― What are your future goals?

At the moment, I have stepped away from the front lines of research and have dedicated myself to management. Managing an organization and cultivating my subordinates is a lot of fun, but if possible, I’d like to do research again (laughs). Even from a manager’s perspective, those engaged in research have this shiny aura about them, and more than anything, I love producing something from nothing.
In addition, since long ago, I have been thinking about how to recreate a human brain through computing. This is the dream I have been chasing. While AI has certainly made impressive progress, it is still nowhere close to recreating the human brain. What’s more, unlike artificial intelligence, humans have emotions. I am always thinking about how to go about recreating such an advanced mechanism. I think it would be wonderful to delve into research again with my subordinates.