Aiming to Create a System for Computers to Acquire Knowledge and Grow Autonomously


Yuji Sato, Professor

Department of Computer Science, Faculty of Computer and Information Sciences, Hosei University

Posted Jan. 10, 2019

Faculty Profile

Professor Yuji Sato is committed to using cutting-edge information processing technology in the service of society. He is currently forging ahead with applied research on intelligent computing, with a focus on “evolutionary computation.”

In pursuit of “evolutionary computation” to create computers that get smarter by themselves

I employ a method known as evolutionary computation to do applied research on intelligent computing (*1), a field of artificial intelligence (AI). The aim is to create a system whereby computers get smarter on their own, just as living things evolve by acquiring wisdom in order to adapt to changes in their environment.

So far, AI research has mainly addressed methods for recall and reuse of massive volumes of stored data (memory). Under these methods new knowledge cannot be acquired unless data is explicitly added by humans.

Today, however, advancements are being made in research to enable computers to acquire new knowledge. In other words, we are exploring methods for computers to develop autonomously.

Evolutionary computation (*2) has the potential for application in many different fields, but one area that my research is currently focused on is resolving the problems of multi-objective optimization.

Multi-objective optimization is a method of finding the optimal answer for multiple objectives simultaneously. When developing a new automobile, for example, needs of different types have to be addressed – such as “faster speeds” together with “larger passenger capacity.” Multi-objective optimization looks at how to address them together.

Previously, developers tried to tackle such challenges by giving greater emphasis to the objectives considered more important. For unfamiliar challenges and projects in new fields, however, it can be difficult even to work out the priorities.

In such cases, employing evolutionary computation enables us to quantify and make predictions regarding multiple options simultaneously. This allows the solution options to be narrowed down and makes it easier to select the optimal goal.

*1. Intelligent computing: Information processing technology that equips computers with human-like intelligence so that they can handle complex problems accommodating inaccuracy, inconsistency, and imperfection.

*2. Evolutionary computation: Living things evolve by generating variations through the processes of procreation and by adapting – or not – to their environment. Taking hints from this biological evolution mechanism, evolutionary computation is a method for seeking effective solutions through computer-based random variation and parallel manipulation to get closer to the desired results.

A New Path Uncovered after an Unexpected Stumbling Block

When I worked in the corporate sector, I was involved in hardware development. I did research and development on neural networks, which are artificial systems that use LSI (semiconductor integrated circuits) to recreate the configuration of neural circuits within the human brain. 

The turning point came when I was posted to a research institute in Tsukuba to participate in a national government project. Having just married, I found it lonely to have to live on my own, but I was determined to make use of the environment in which I could devote myself to research, produce a dissertation, and earn an academic degree.

But just three months after I moved to Tsukuba, something completely unexpected happened. The budget for the planned research project was not approved, and I was ordered to change direction and begin work in a new field of software development.

I was plunged into darkness, wondering what all my efforts up to that point had been worth. Full of uncertainties about the future, I eventually decided that I would work on information processing related to biological evolution and the capabilities of the human brain. With no choice but to press forward, I immersed myself in this research, and finally obtained the degree that I had been hoping for. A presentation on my findings at an international conference led to the opportunity to come to Hosei University. I have no regrets about my choices.

It is important for researchers in evolutionary computation not only to have computer-related knowledge and skills, but also to incorporate insights from disciplines such as brain science, biology, and genetics. Frontier research fields like this, which span multiple disciplines, are precisely the ones where the needs are greatest, and I believe that as we solve problems in these fields we will be able to produce outcomes that are useful in real life.

“Practical Wisdom” Means Furnishing Society with Ideas Fostered in Hosei’s Free-Thinking Environment

I feel that Hosei is a university where you can do research in an atmosphere of freedom, just as envisioned by the university’s ideal of “freedom and progress.” My personal wish is to put the ideas gained in this free-thinking environment to use in real life: the real meaning of my research lies in contributing to society. That is what “practical wisdom for freedom” means to me.

I try to foster individual autonomy and self-motivation in my students. Recently, possibly due to experiences such as studying for entrance exams, more and more students seem to have a passive mindset, responding based on what they have memorized rather than thinking for themselves. They immediately want to know the answer when they get stuck, but I encourage them to think it through on their own wherever possible. When they go out into wider society, they will need to have the capacity to think logically and come up with solutions themselves. I believe that you can only gain this capacity through a process of trial and error, struggling and finally arriving at your own solution.

Computers can evolve by repeating the same process over and over, and accumulating experiences of success. I expect students, too, to have a variety of experiences, acquire new insights autonomously, and grow even more as individuals.

Practical Wisdom in Action: Challenges in Applying Evolutionary Computation in Practice

I am currently working on a number of problems with a view to utilizing evolutionary computation more practically.

One of these is improving processing speed. When seeking solutions in multi-objective optimization in particular, the more different goals there are to consider, and the more complex the challenge is, the more processes are required and the longer it takes to produce results. Computation speed can easily be increased by using a large-scale computer, but this requires expenditure on maintenance and physical space for installation, which restricts the possibilities for use to certain fields. If processing efficiency could be raised on more accessible personal computers, the scope of possible uses would expand dramatically.

With this in mind, I honed in on the topic of parallel computation utilizing a GPU (graphics processing unit). A GPU is a device specializing in processing functions related to image displays, and is a component in personal computers. Making effective use of this device is expected to enable more low-cost, high-speed processing.

Moreover, evolutionary computation will improve the capacity to derive accurate solutions. If there is a physical fault somewhere in the system, the computer is able to avoid it automatically and yield a solution without halting the process. This improves the sustainability of application programs.

I believe that this kind of technology that enables reliable high-speed processing in limited space could be deployed in many different settings. I hope that in the future it will prove useful in automobiles, medical and other precision instruments, and similar fields where high performance is essential.

Ever since I was involved in hardware development, I have been trying to help shed light on human brain functions. But there is still so much we don’t understand. This is why it’s so intriguing: I never lose interest in the question of how we can create systems and hardware that are as close as possible to the human brain.

Yuji Sato, Professor

Department of Computer Science, Faculty of Computer and Information Sciences, Hosei University

Born in Tokyo in 1957.

Graduated from the Department of Applied Physics, School of Engineering at the University of Tokyo, and joined Hitachi, Ltd. After working in Hitachi’s Central Research Laboratory, became an Associate Professor in the Faculty of Computer and Information Sciences at Hosei University in April 2000. Professor since April 2001. Was a visiting researcher at the University of Illinois at Urbana–Champaign (IlliGAL) for one year from September 2007. Received a Highly Commended Paper Award for the International Journal of lntelligent Computing and Cybernetics in 2015. Member of the Information Processing Society of Japan, the Japanese Society for Evolutionary Computation, ACM/SIGEVO, and IEEE. Holds a PhD in engineering.