Deep Learning Platform with World-Class Speed, AI Services that Support
Industries and Operations
with “Zinrai,” Fujitsu’s AI framework
today announced it has developed five Zinrai-related services that are being
rolled out incrementally in Japan, helping customers accelerate their use of
artificial intelligence. These services are based on the company’s foundation
of AI technology and knowledge, and are part of the “Human Centric AI
Zinrai” framework, Fujitsu released November 2015. Fujitsu is positioning
these services on its FUJITSU Digital Business Platform MetaArc, which works to
accelerate the business innovation of customers.
The services being introduced are FUJITSU AI
Solution Zinrai Platform Service, which offers 30 AI functions in API(1) form,
developed in over 300 AI-related projects and field trials; FUJITSU AI Solution
Zinrai Deep Learning, a deep-learning platform service that implements the
world’s fastest class of deep-learning processing; and FUJITSU AI Solution
Zinrai Consulting Service, FUJITSU AI Solution Zinrai Integration Service, and
FUJITSU AI Solution Zinrai Operations Service, for total support to customers
on AI, from consulting to deployment and operations.
Zinrai Platform Service and Zinrai Deep Learning
are combined with consulting, deployment, and operations services so that
customers can quickly build high-quality, high-performance AI-based business
Fujitsu regards AI as a core technology that can
accelerate digital innovation, and continues to develop technologies in the
field. Also, by applying the cutting-edge technologies it has accrued by
developing supercomputers, the company aims to push technological
transformations even further, helping customers to transform their existing
operations and create new businesses.
As companies have come to embrace new ways of
using data to expand their businesses and transform their operations, they have
grown more interested in AI as technology critical to streamlining and
accelerating that process.
Figure 1: Positioning of AI within MetaArc
Fujitsu has been working on AI since the 1980s,
mostly by Fujitsu Laboratories Ltd., and in November 2015 organized its AI
knowledge and technology under the Zinrai framework. Having many of Japan’s
leading companies as customers, Fujitsu has studied and tested the application
of AI to business systems together with more than 300 customer firms. The
services being announced represent the technologies and know-how that were
found through that process to be especially important to customers using AI.
About the New Services
Figure 2: Service framework
1. Zinrai Platform Service, providing practical
AI technologies via APIs
Zinrai Platform Service offers a set of highly
practical AI functions proven through more than 300 AI-related business
projects and field trials, made available as a service in the form of APIs. A
learning-model build function allows customers to easily make learning models
appropriate to their operations, rather than just being able to use pre-built
models. For customers that require a deep-learning processing platform, this
service provides Zinrai Deep Learning, which boasts world-class speeds.
Fujitsu Platform Service provides a total of 30
APIs, comprised of 21 function-oriented APIs, divided into three categories –
Sensing & Recognition, Knowledge Processing, and Decision & Support –
of building-block technologies, and nine application-oriented APIs, which
combine building-block technologies by use case and include functions and
knowledge to facilitate the use of AI with customer operations. Customers can
quickly put together an AI-powered system by choosing their required APIs.
The first offering, to be provided from April
2017, includes 7 function-oriented APIs, such as for image recognition, speech
to text, and knowledge searching, as well as two application-oriented APIs,
such as demand prediction, with the full range expanding to 30 types of API
before April 2018.
Figure 3: APIs provided with Zinrai Platform
2. Zinrai Deep Learning, providing the world’s
fastest class of learning capacity
As part of Zinrai Platform Service, Fujitsu is
providing the deep-learning platform service Zinrai Deep Learning, which offers
the world’s fastest class of learning capacity, using software technology that
implements high-speed deep learning(2) and parallel-processing supercomputer
technology developed by Fujitsu Laboratories, in conjunction with the Tesla
P100, the latest GPU(3) from NVIDIA Corporation of the US. Based on this
service, customers can combine APIs best meeting their needs, to implement
high-speed, high-quality, AI-based systems.
3. Support services for everything from
consulting to deployment and operations of AI systems
To fully support customers’ use of AI, from
planning to deployment and operations, Fujitsu is providing Zinrai Consulting
Service, with which Fujitsu’s AI specialists define the best scenarios for
using AI based on a customer’s business issues and needs; Zinrai Integration
Service, to help customers quickly plan and set up AI-powered systems; and
Zinrai Operations Service, for maintenance of learning models post-deployment.
As the range of applications for AI grows
broader, Fujitsu plans to continue implementing building-block AI technologies
that Fujitsu Laboratories has taken the lead to develop as APIs. To strengthen
the use of AI in edge devices, such as communications devices, robots, and
automobiles, the company is equipping Zinrai Platform Service with
learning-model distribution functions for such devices.
Fujitsu also plans to offer the hardware,
software, and services of Zinrai as on-premise packages to customers planning
on building their own in-house AI-powered systems.
In addition, Fujitsu is developing a proprietary
DLU (4) processor specifically for deep learning, using processor technologies
pioneered on the K computer(5) and advanced CMOS(6) technology. This is due to
begin shipping in fiscal 2018.
Yutaka Matsuo, Project Associate Professor,
Institute of Engineering Innovation, University of Tokyo
Deep learning is the result of 50 years of
breakthroughs in AI, but will bring extraordinary social changes, on the same
order as the Internet. My hope is that the DLU dedicated deep-learning
processor will increase Japan’s industrial competitiveness at the same time
that it contributes to the world.
Masataka Osaki, Japan Country Manager, VP
Worldwide Field Operations, NVIDIA Japan
Deep learning is the driving force behind the use
of AI, which is referred to as the intellectual industrial revolution. Deep
learning opens the door to a vast range of possibilities, but it also requires
huge amounts of computational power. Combining NVIDIA’s newest, high-end GPU,
the Tesla P100, with Fujitsu’s supercomputer parallelization technologies
enables construction of the world’s fastest learning environment for deep
learning, accelerating the development of high-accuracy models with large-scale
neural networks, which will contribute to the creation of new services and
products from both our companies.
Kazunori Inari, Vice President of Operations,
Research Institute of Systems Planning, Inc.
Together with the Department of Biomedical Data
Intelligence, Graduate School of Medicine, Kyoto University (Professor Yasushi
Okuno), our company began research on April 1, 2016, on the theme of
“predicting chemical-compound activation using deep learning,” and
had the opportunity to participate in early access to Zinrai Deep Learning.
Using the powerful, multi-GPU environment of Zinrai Deep learning and our own
framework for high-speed data feeds, we were able to greatly accelerate our
research, the results of which were presented at the 2016 meeting of the
Chem-Bio Informatics Society on October 25, 2016. Research Institute of Systems
Planning has built a robust partnership with Fujitsu for applying AI
technologies to a range of industries, even as we have made progress together
with Professor Okuno’s lab.
Pricing and Availability
Application Programming Interface. A protocol
stipulating data formats and procedures allowing the functions and data handled
by one computer program to be called externally by another program.
(2) Software technology that implements
high-speed deep learning
See “Fujitsu Develops High-Speed Technology
to Process Deep Learning,” press release dated August 9, 2016 (http://www.fujitsu.com/global/about/resources/news/press-releases/2016/0809-01.html).
Graphics Processing Unit. A processor specially
designed to process images; now often used for other purposes thanks to their
very high processing speed.
Deep Learning Unit.
(5) K computer
A supercomputer and the core system of Japan’s
Ministry of Education, Culture, Sports, Science and Technology’s High
Performance Computing Initiative (HPCI), developed jointly by Fujitsu and
RIKEN. When it went into general service in 2012, it had an estimated
performance in the 10-petaflops range.
Complementary Metal-Oxide Semiconductor. A
semiconductor characterized by very low power consumption because nearly no
current flows when it is in a static state. Used in the CPUs for Fujitsu’s
mainframes and supercomputers.ax