Fujitsu has delivered its RAIDEN (Riken AIp Deep learning ENvironment) computing system for AI research to the RIKEN Center for Advanced Intelligence Project (hereinafter “AIP Center”), the artificial intelligence (AI) research base of RIKEN (Japan's largest comprehensive research institution). RAIDEN, which has been used for R&D since April 2017, makes use of all the supercomputing expertise accumulated by Fujitsu over the decades, to support the AIP Center’s R&D efforts. We asked director Masashi Sugiyama and Kazuki Yoshizoe, a unit leader, to talk about the present and future of the RAIDEN system and the AIP Center.
[ Published: 2017/10/27 ]
[ Updated: 2018/4/20 ]
|Total theoretical computing performance||54 PFLOPS (FP16)|
NVIDIA DGX-1 54 units
FUJITSU Server PRIMERGY CX2550 M4 64 units
FUJITSU Server PRIMERGY RX2530 M2 32 units
FUJITSU Server PRIMERGY RX2540 M2 6 units
FUJITSU Storage ETERNUS DX200 S3 8 units
FUJITSU Storage ETERNUS DX100 S3 1 unit
FUJITSU Software FEFS
Univa Grid Engine
The RAIDEN system is comprised of two kinds of calculation server, utilizing the latest CPUs and GPUs, and a storage system. In addition to standard deep learning software for DGX-1, Fujitsu also developed a customized software environment that can be used within an on-site, secure network. Furthermore, the system boasts simple and flexible operation and management functions (for creating and reproducing computational environments), and a level of security and reliability suitable for highly sensitive data, such as personal information and intellectual property information.
AI (artificial intelligence) is where many people are pinning their hopes for a technological breakthrough that will transform society from its current state to its next stage of evolution. Right now, all over the world, the competition in AI-related research is intensifying.
Japan has been slow to get started in AI, but in 2016, the Ministry of Education, Culture, Sports, Science and Technology (MEXT) made a serious push with the launch of the Advanced Integrated Intelligence Platform project (AIP project). The AIP Center was then established as the key research base of the project. AIP Center Director, Dr. Masashi Sugiyama, commented as follows:
Masashi Sugiyama (D.Eng.)
RIKEN Center for Advanced Intelligence Project (AIP Center)
“Previously, individual researchers pursued research on their own, but it’s hard to make any big leaps this way. Our hope is that with enough funding from the government we can assemble a wide variety of researchers. If we combine multiple kinds of research work and bring researchers into close contact with each other, I’m sure we can be effective.”
Dr. Sugiyama highlights, however, that the intention is not to establish a “Japan only” center; researchers will be assembled from all around the world. He says, “We want to become an international research center. For AI research, it is certainly not enough to have only Japanese researchers. I would go as far as to say that having an international mix of researchers is an essential condition for the success of the center.”
At the moment, the big trend in the AI field is deep learning. Particularly in image and speech recognition, the use of deep learning has driven performance to a much higher level, so a large part of the research has shifted towards deep learning.
The U.S. has become the world leader in the AI field. In applied deep learning-related research especially, the U.S. has surpassed other countries with the overwhelming scale of its investment. “On the other hand,” notes Dr. Sugiyama, “so much of the theory remains unknown, so I think that at least in research on basic theory, AI is still a field in which Japan has a good chance to make its presence felt.”
The AIP Center is composed of three main research groups. Firstly, there is the Generic Technology Research Group, which deals with fundamental research. “The AIP project was granted a 10-year research period by the MEXT. To get such a long research period is rare, anywhere in the world, and I think that many overseas researchers are very attracted by this.”
The second group is the Goal-Oriented Technology Research Group. “It develops technology focused on specific applications. Since American IT companies are investing billions of dollars in this field, it’s hard to compete with them on a level playing field. Consequently, we are thinking of making use of AI to develop solutions to the social issues unique to Japan, such as medical diagnosis, or urban infrastructure management. One advantage we have in tackling these kinds of challenges is that we can utilize the huge wealth of accumulated public data available in Japan.”
The third group is the Artificial Intelligence in Society Research Group, which studies the social impact of AI. “Many researchers from the humanities participate in this field. Culture and systems differ from one country to another, so the group also tackles comparative research.”
Kazuki Yoshizoe (Ph.D.)
Unit Leader, Search and Parallel Computing Unit
RIKEN Center for Advanced Intelligence Project (AIP Center)
The RAIDEN computational system supplied by Fujitsu plays an active role in all this research work. Dr. Kazuki Yoshizoe, an AIP Center researcher (leader of the Search and Parallel Computing Unit) and the person in charge of introducing the system, reflects as follows. “The key point was the computational speed and convenience. In research, victory or defeat hinges on how quickly or how efficiently you can give shape to an idea, so that’s what we gave importance to.”
Often with conventional supercomputers, a software upgrade would make previous research results unusable, and considerable effort was required just to get started again. “For this reason, we adopted a technology called ‘containers’ for storing past research assets for each software. The fact that Fujitsu was providing the system—a company with a comprehensive range of leading-edge IT technology, including supercomputers—gave us confidence. At the same time, we can use a large number of GPUs to ensure excellent computational speed. At this time, only a small number of vendors around the world are capable of putting together a system like this, and Fujitsu is one of them.” (Dr. Yoshizoe)
Currently, some of the areas that make intensive use of the RAIDEN system include research to support diagnosis and treatment by means of medical image recognition and analysis, and research to assist with the maintenance of infrastructure, involving the use of drones to capture images of bridges and other structures. As an example, the video images captured by drones constitute a huge volume of data, so the massive storage capacity of RAIDEN helps considerably.
“Some researchers have said that they consider RAIDEN a lifeline for their work. Already, RAIDEN has become an indispensable tool for the center’s research.” (Dr. Yoshizoe)
“In terms of the supercomputer, I feel that collaborating with Fujitsu, a company with outstanding know-how within Japan, worked out very well for us. It’s not that you can immediately start using a supercomputer after you buy it; the vendor needs to take care of its environment and various other things too.” (Dr. Sugiyama)
RIKEN and Fujitsu have set up the RIKEN AIP-FUJITSU Collaboration Center, as part of a multifaceted cooperative relationship. Together, they will pursue joint research in the field of AI, as well as work together on the development of the post-K supercomputer—a successor to the K supercomputer. In addition, Fujitsu is actively engaging in the AI field, by systematizing AI-related knowledge and technology, in the form of Zinrai, or “Human Centric AI.” RIKEN also expects that its relationship with Fujitsu will not be limited to RAIDEN, but rather that it will grow and broaden.
“What we are trying to do is pure research, but we also have a strong desire to make use of the results in the real world. That is, we want to turn our algorithms into useful products and get them implemented around the world. However, a research institution cannot do this on its own. In practice, this goal can be achieved only through cooperation with a company that has a market it is actively engaged in. So we have high hopes for our partnership with Fujitsu and so on.” (Dr. Sugiyama)
Benefits in the area of human resources should not be overlooked either. In Japan, the number of graduate students pursuing doctoral programs remains low, and Dr. Sugiyama feels a sense of crisis, believing that the foundations of Japanese science will be threatened if the number of doctoral-level personnel does not increase. “We are conducting research together with people from Fujitsu. It will be very positive if we can provide more opportunities for company employees who are inclined to engage in academic research at AIP Center, even as they remain with the company,” said Dr. Sugiyama.
Advances in AI research happen very quickly, so within a few years the current system will probably need to be renewed. AIP Center is hoping that Fujitsu will be able to respond appropriately to this change.
Masaaki Sato, Business Strategy & Development Division
Rie Murakami, Sales Division
Tomohisa Yazu, Computational Science and Engineering Solution Division
Technical Computing Solution Unit
Even for Fujitsu, RAIDEN was a technologically innovative system, but through cooperation with other Fujitsu Group companies and partners, we were able to satisfy AIP Center’s request for “a computing environment that enables researchers to concentrate on research.”
Looking ahead, we would feel honored if the multifaceted support we provide to AIP Center, in relation to leading-edge supercomputing technology, Zinrai, and other AI-related R&D, allows us to create a new tomorrow together.
From the left Masaaki Sato, Rie Murakami, Tomohisa Yazu.
From the left Kazuki Yoshizoe, Masashi Sugiyama of RIKEN Center for Advanced Intelligence Project (AIP Center).
|Headquarters||2-1 Hirosawa, Wako, Saitama 351-0198, Japan|
|Found||on March 20, 1917|
The titles, numbers, proper nouns are information of the posted date. Please understand that there is a possibility that changes have been made at the time of viewing.
Share this page