Interview Vol. 003

My area of expertise is black-box optimization, focusing on the algorithmic research of evolutionary computation and reinforcement learning. Optimization is the problem of finding the optimal solution that minimizes or maximizes the value of the objective function that represents the solution’s goodness. Black box optimization is a problem in which the objective function is not explicitly given but is given by a simulator or other means.

Usually, mathematical assumptions such as differentiability, linearity, and convexity of the objective function are used when solving optimization problems, but these are difficult to achieve, and the subjects are problems that could previously only be solved by human experts through trial and error. While deep learning and machine learning algorithms used in recent chat GPT and AI require vast amounts of teacher data, evolutionary computation, and reinforcement learning are characterized by the ability to collect data through trial and error on their own and find superior solutions if a simulator is available, which allows them to solve problems. Therefore, it is possible to find solutions that would have been difficult for humans. For example, the current Shinkansen bullet train has a leading shape designed to reduce noise when exiting a tunnel, but it is difficult for humans, even experts, to design such a shape. In fact, this shape was discovered through trial and error on a computer using evolutionary computation. Even if the answer is found in a hundred years, the times will already have changed, so I am researching to create an efficient algorithm that will make it effective and not waste trial and error.

It all started in my third year of undergraduate studies at my part-time job, where I needed to find the shortest path given multiple points. It was called the traveling sales problem, and there happened to be a lecture in my university class that introduced an algorithm for evolutionary computation to solve it. I was so impressed that I joined the laboratory of the professor in the Department of Intelligence Science at that time.

No, it is actually an event company. Using lasers to draw pictures or produce text is often done at concerts and other events, isn’t it? By chance, I received a request to optimize how the laser pictures and text are displayed on that laser display.

Yes, of course. I move that X-axis and Y-axis mirror and reflect it there to paint the picture. Even if the mirror is moved at high speed, the number of pictures that can be painted with a single piece of equipment depends on whether or not it can be connected by the shortest path. To paint a large number of large pictures, we must connect them in the shortest path. That is how I was attracted to the fun and usefulness of optimization.

It all started when Prof. Miyake, Director of the Center of Data Science and Artificial Intelligence, asked me to help him with his work at the Center. The predecessor of the Center already existed at that time, but Prof. Miyake explained to me about the activities of the Center, and I said, “Yes, I am willing to work with them.” and I joined the Center. It is actually very rewarding to be involved. They have just started up the Center, but they are trying to start new things one after another. The DS&AI TF program, which is one of today’s themes, is one of such things. They are trying various new things after establishing the Center, so it is very difficult, but I feel that it is very rewarding.

This is part of the Ministry of Education’s plan to expand the number of people capable of DS&AI throughout Japan to 500,000/year and is currently being tried at universities nationwide. In the Kanto region, there are three core schools, the University of Tokyo, the University of Tsukuba, and the Tokyo Institute of Technology, and we are proceeding in the way of expanding in each region with such schools as core. In particular, the Tokyo Institute of Technology started its education from the graduate school, and from there, the program was extended down to the bachelor’s program. Therefore, it is a very unique university. It started at the expert level in graduate school and went down from there.

It develops in the form of an educational program. As the basic concept of education at the Tokyo Institute of Technology’s Center of DS&AI, it offers a consistent educational program from the literacy level of first-year undergraduate students to students in all six academic fields in the form of broad-based specialization. From this year, we have begun offering Expert Level Plus, a specialized education that goes beyond the Expert Level.

Synergies can be created by teaching DS&AI further to students with their expertise. We believe that the good thing is that it can contribute to an extensive range of research. The program’s design is to have students study basic knowledge in basic courses in the first and second year of undergraduate and then study specialized DS&AI in graduate school.

I am the Vice Director of the Center of DS&AI and assist Prof. Miyake, the Director of the Center. In addition, I am also the Director of the University-wide Education Department, one of the four Departments (University-wide Education Department, Social Partnership Department, Information Infrastructure and Public Relations Department, and Planning and Research Department). I am working with various teachers to design the curriculum and teaching materials for the entire institute. TF is also connected to that.

The human resource development philosophy of the Center of DS&AI has three pillars: First, learning various methods by making full use of DS&AI. Next, with DS&AI as a common language, we will conduct research and development while communicating with experts in various fields. Finally, the idea is to develop top talent who can teach DS&AI. TF is this last part. We have created an educational program with more than 40 partner companies and have their researchers and engineers give lectures. In addition, from this year we are launching an internship course and a TF program to develop teaching skills through TAs and other educational experiences. In Japan, there is a lack of human resources who can teach DS&AI, so we would like people to learn here so that they can teach DS&AI at universities and companies in the future.

We have a three-step gate as the mechanism, starting with BTA (Basic Teaching Assistant), who is a beginner in DS&AI, then aiming to ATA (Advanced Teaching Assistant), and finally TF (Teaching Fellow). It is designed for students graduating with a master’s degree to ATA and for doctoral students to work hard until TF. BTA is at the introductory level, and you can become a BTA by studying the guidebook and other information on teaching, then doing a TA, submitting a report, and being certified. Of course, expertise in DS&AI is required. Each unit is pointized to make it easy to understand and proceed. For example, to become an ATA, you must accumulate 28 points worth of TA, submit a report, and then gather at a TA work reporting meeting to talk about your TA experience and discuss what innovations you made and what you could do to teach better. Accumulating points means attending lectures and exercises and teaching with the teacher, so we want students to be able to put themselves in the teacher’s position, watch the class, and steal teaching skills from it. The highest level of TA is the TF. Through lectures offered in cutting-edge data science and other courses offered at the Expert Level Plus, they will gain further specialized high knowledge of AI technology, or AI ethics and AI in society. In addition, they will have to create teaching materials and even do a mock class. If they pass all of these, they can become TFs. The advantage of authorizing your teaching experience this way is that you can adequately state on your resume that you have done this.

As for skills, a complete and in-depth understanding of the field is required as the basic prerequisite for teaching. In addition, the baseline is to have a deep understanding of the expert level or expert level plus specialized DS&AI knowledge of the programs we are working on and the ability to operationalize that knowledge in research and other areas.

Furthermore, the students who gather in the classroom range from beginners to those who are somewhat proficient, but there is only one teacher, so it is very important to assist the students with how to pick them up. It is important to have TAs who can find students who might be left behind and proactively stay close to them to show them how interesting they are. The ideal TA is one who can give students the feeling of learning to use the system properly rather than just thinking, “I submitted a report and got a grade, so that is good enough.” In the end, it is a feeling that what is required is a part of humanity. We started last year and already have a total of 40 to 50 TAs.

It has been more than twenty years, so my memory of thinking about going to the Tokyo Institute of Technology has faded considerably, but yes, I did like science, or rather, mathematics and physics. As for computers, I had been programming and building my electronic circuits before entering university. Also, my high school mathematics teacher was from the Tokyo Institute of Technology. He said that if you like that kind of thing and want to do research, the Tokyo Institute of Technology is a great place to go. He recommended that I go there, and I followed his advice. I remember that he was studying something related to nuclear reactors at the Tokyo Institute of Technology. He was not just a mathematics teacher.