Fujitsu launched commercial services for Digital Annealer on May 15, 2018. Participants in the discussion during the frontline session at Fujitsu Forum 2018 included Professor Ali Sheikholeslami of the University of Toronto, Canada and Mr. Andrew Fursman, CEO of 1QB Information Technologies Inc. (1QBit). During the session entitled the "New Future Opened Up by Digital Annealer," we introduced how Digital Annealer is used on the front lines of medical care and other areas closely related to people's lives.
[Fujitsu Forum 2018 Frontline Session Report]
Corporate Executive Officer
Fujitsu Laboratories Director Takeshi Horie, who served as session moderator, introduced Naoko Yoshizawa, Fujitsu's Corporate Executive Officer in charge of the Digital Annealer business.
Yoshizawa opened the session with these words: "The three technologies of HPC, Deep Learning, and Digital Annealer accelerate AI." She explained that in the HPC arena, Fujitsu developed the K computer and is now developing its successor. With respect to Deep Learning, she said that Fujitsu will release a dedicated processor (DLU) in 2018. Yoshizawa remarked, "Only Fujitsu has all three of these technologies, including Digital Annealer, and each element correlates with the others when building solutions." She then explained Digital Annealer's background and future roadmap.
Digital Annealer Bridges Today's and Tomorrow's Computing
Amid expectations that practical use of quantum computers will bring about the next generation of computing, the development competition among major players worldwide is intensifying. Despite this, it is thought that it will take more time before quantum computers that can solve diverse problems in the real world can be put into practical use.
Given these circumstances, Fujitsu has developed a completely new type of computer dedicated to solving combinatorial optimization problems, Digital Annealer, which uses a digital circuit design inspired by quantum phenomena. On May 15, 2018, Fujitsu launched a cloud service with Digital Annealer. Combinatorial optimization problems can be found in a variety of sectors, including drug discovery, finance, and distribution; solving such problems offers many business opportunities.
Digital Annealer uses a digital circuit design inspired by quantum phenomena.
Since Digital Annealer is built using existing semiconductor technology, it operates stably at normal temperatures without requiring a special cooling system. Digital Annealer has demonstrated a great ability to practically solve various combinatorial optimization problems which require large-scale computation and high accuracy. In Fujitsu's in-house usage, Digital Annealer optimized movements for picking up parts stored in a warehouse at a production site, successfully reducing transfer distances by 20%. By optimizing the locations of parts inside the warehouse, Digital Annealer can further reduce movement distances by up to 45%.
Example of an in-house practice at Fujitsu IT Products Limited: Improving productivity by optimizing line flow inside a plant (Japanese)
Quickly solving combinatorial optimization problems that previously took an extremely long time to calculate
We will begin by providing the first generation of Digital Annealer as a cloud service. Later, we plan to provide the second generation Digital Annealer UNIT (DAU) with improved performance as well as Digital Annealer for on-premises usage in the third quarter of fiscal 2018. Also, we are currently developing a large-scale parallel processing technology for DAUs. With these technologies, we will dramatically increase the scale of our capabilities in order to handle 1 million-bit problems by fiscal 2019.
The 1,024-bit first generation of Digital Annealer is initially available as a cloud service, an up to 8,192-bit DAU will appear within this year, and then on-premises products will become available.
The second generation of the DAU will have improved accuracy and increased scale. We think that application areas will expand by adding the ability to change the configuration to address the characteristics of the problem at hand: some problems require high accuracy (e.g., financial problems), while others require large-scale computation (e.g., chemical problems).
Expanding application areas by switching the configuration
In response to Yoshizawa's explanation, Fujitsu Laboratories Director Takeshi Horie summarized the discussion: "We have already conducted verification of Digital Annealer with customers. Along with technology development, our cloud service and technical service are gradually being deployed globally, advancing our efforts to solve social problems."
Combining 1QBit Software with Digital Annealer Is the Best Solution to Solve Optimization Problems
1QB Information Technologies, Inc.
Next, as Fujitsu's partner for globally deploying Digital Annealer, Horie introduced Mr. Andrew Fursman, CEO of Canada's 1QBit, which is the world's top quantum computing software vendor.
Fujitsu has partnered with 1QBit, the world's top software vendor for quantum computing. By collaborating with 1QBit, Fujitsu will provide Digital Annealer services and develop business globally.
Partnership with 1QBit
Mr. Fursman introduced a case study about solving the following combinatorial optimization problem with Digital Annealer: "If we try all the combinations of 250 lights on-stage to obtain the optimal combination for light distribution, the number of combinations is 2 raised to the power of 250. This number is nearly the number of observable atoms in the universe. To try all those combinations, in theory it will take more time than the history of the universe."
Another example can be found in the financial industry. JP Morgan developed a risk parity portfolio using the hierarchical type of the minimum variance method. The company proved that risks can be reduced 20% by calculating the correlation of combinations with stock prices.
In the chemical field, the similarity of molecules can be analyzed in terms of numeric values by turning molecular structures into mathematical graphs. Digital Annealer's great advantage is its ability to analyze graph structures, and it will become able to perform more complex analysis in the future.
Such examples can also be found in other fields; the innovation brought about by this new type of computing has only begun. Mr. Fursman summarized his thoughts: "When processing capacity increases explosively in the future, I want to realize innovation in this new type of computing through collaboration between Fujitsu and 1QBit."
Horie then asked a question: "1QBit is the top quantum computing software vendor. What led you to form a global partnership with Fujitsu?" Mr. Fursman responded, "Fujitsu's Digital Annealer serves as a bridge with future quantum computers; at the moment, it is the best solution to run 1QBit's software."
The University of Toronto Working on Interdisciplinary Problem Solving with Digital Annealer
Professor Ali Sheikholeslami
University of Toronto
Finally, Horie introduced Professor Ali Sheikholeslami of the University of Toronto, Canada, who has been conducting joint research with Fujitsu Laboratories. The University of Toronto is a first-class research institution focusing on AI and quantum computers. In March 2018, Fujitsu Laboratories established the Fujitsu Co-Creation Research Laboratory at the University of Toronto, a joint research hub. At this hub, Fujitsu has been conducting joint research with several university departments and researchers in an attempt to expand Digital Annealer's areas of application into multiple fields, including smart transportation, finance, networking, and medical care.
Since joining Fujitsu Laboratories as an intern during his Ph.D. studies (1988), Professor Sheikholeslami has been engaged in joint research activities in various fields with Fujitsu. Professor Sheikholeslami has been conducting research on Digital Annealer at the Univesity of Toronto over the past few years.
Partnership with the University of Toronto
Application fields of Digital Annealer include radiation therapy, brain-machine interfaces, urban transportation, wireless applications and connectivity, and economics. These fields may appear unrelated, but they all share a common need for optimization.
Cancer is the leading cause of mortality worldwide. In 2012, 14 million people developed cancer, and in 2017, 80,000 died of cancer in Canada, while 378,000 died of cancer in Japan. Research on cancer treatments is underway in many locations.
The photograph below shows a volumetric modulated radiation therapy (VMAT) machine, which is used all over the world. VMAT is a cancer therapy machine that rotates around the patient, reshaping and adjusting the intensity of the radiation beam. In such treatment, it is important to optimize the applied radiation in terms of radiation angle, area, and period in order to do maximum damage to cancer cells while minimizing damage to the surrounding normal cells.
Optimizing radiation to maximize the radiation applied to cancer cells while minimizing the impact on the surrounding normal cells is important in cancer radiation therapy.
The University of Toronto has been conducting joint research with state-of-the-art hospitals to use Digital Annealer to find the optimal solution in under 30 minutes. In an attempt to respond to cancer organs with complex shapes in the future, the university is also conducting research to use multiple DAUs to obtain the optimal solution.
We will now introduce other application examples. Dr. Taufik Valiante has been conducting research on brain structure analysis. He does research that uses Digital Annealer to analyze how brain neurons are linked, the characteristics of individual neurons, and the connections among neurons. We believe that the results of such analysis can be used to treat brain-related illnesses, such as Alzheimer's disease and depression.
The next two examples introduce use of Digital Annealer in relation to urban transportation, a social issue.
Professor Alberto Leon-Garcia has been researching urban transportation for many years. Using Digital Annealer enables him to find the optimal solution in real time in response to ever-changing road conditions. This is expected to provide unlimited ability to reduce traffic jams.
In the near future, connected cars will spread. Connected cars communicate with infrastructure and networks as well as other vehicles. Professor Shahrokh Valaee has been using Digital Annealer to research optimized use of limited wireless resources.
Fintech is another important theme, as also mentioned by Mr. Fursman. Professor Yuri Lawryshyn has been studying this area with respect to investments that minimize risks, algorithms for high-speed transactions, and credit assessment at the University of Toronto. He is now utilizing Digital Annealer to accelerate solutions in these areas.
As explained above, while Digital Annealer is based on conventional general-purpose computing, its excellent ability to solve combinatorial optimization problems is comparable to the quantum annealing method. It is extremely user friendly and achieves a practical level of accuracy.
In response to these remarks by Professor Sheikholeslami, Horie summarized them as follows: "The University of Toronto and Fujitsu Laboratories have a long-term relationship that spans 20 years. I think you now can see that the University of Toronto is seeking out various new applications of Digital Annealer and such efforts will produce significant results that greatly impact society."
Further Development through Multidisciplinary Initiatives
Fujitsu Laboratories Ltd.
In response to the discussions, moderator Horie asked: "People in physics, engineering, and applied studies at the University of Toronto unite to effectively work to solve social issues--where does this strength come from?" Professor Sheikholeslami replied: "At the University of Toronto, we have an atmosphere that encourages different departments to study together. In the future, I believe professors and researchers in a variety of fields will use Digital Annealer to achieve various outcomes."
Horie next asked all three participants: "When we seek to achieve the impossible and work on social issues, the key is how far we can get into customers' issues and solve problems with highly specialized expertise. What should be done to realize this?"
Mr. Fursman responded: "At 1QBit, we emphasize the importance of commitment to clients from the early stages of a project. Although we specialize in quantum-related technologies, no one understands the relevant social issue better than the client working on that issue. We collaborate with the client to combine the strengths of interdisciplinary experts and other specialists in particular fields so that we can find the best method to use and apply our technologies. Identifying the areas of needs and matching their solutions is how we do business."
Professor Sheikholeslami remarked: "A startup is one means to solve such problems. We encourage some startups to use Digital Annealer in their fields of interest. I believe in the future a variety of solutions will be brought about when professors and researchers in diverse fields create startups that use Digital Annealer."
Yoshizawa commented: "This issue is relevant not only to Digital Annealer but to AI as a whole. Since AI and Digital Annealer are technologies, utilizing them to solve problems in the truest sense is difficult unless we have knowledge of the relevant sector or domain. Issues that Digital Annealer should solve exist in customers' core business areas, so their business expertise is required. I explore different types of collaboration with diverse parties by taking into account the ecosystem."
Collaboration to Accelerate Digital Annealer and AI
Next, moderator Horie asked for opinions about the relationship between Digital Annealer and AI.
Professor Sheikholeslami replied: "What AI and Digital Annealer can do have many similarities. In machine learning, neural networks are optimized to solve problems. This means that neural networks are adjusted according to the problem to obtain an appropriate solution among the possible solutions. Also, by using Digital Annealer to train machine learning models, we can accelerate processing and reduce the time required for learning."
Mr. Fursman responded: "I am passionate about the Digital Annealer project. We have been studying the concept of 'what can be done to create a machine that thinks' for decades. However, we did not have the arithmetical capacity to realize this. I think Deep Learning is a shortcut for solving issues in AI, but it is not enough. Digital Annealer has the potential to raise AI's potential in an astonishing way and we expect progress in this regard."
Finally, Horie asked the three participants about their expectations and suggestions for Digital Annealer.
Professor Sheikholeslami noted: "I think there are two based on the general direction of our research. First, exploring means to make the hardware of Digital Annealer more flexible. Second, exploring means to re-configure Digital Annealer based on the application."
He further noted: "I believe more interdisciplinary studies are required in the applications area. For example, as I introduced initiatives by five professors, it is important that many researchers and students actually collaborate to expand the application areas for Digital Annealer."
Mr. Fursman answered: "I think the important thing is to make Digital Annealer available to many people. From the viewpoint of software, we at 1QBit will also work on the issue of how we should expand the hardware."
Yoshizawa concluded the session with these words: "As hardware, software, and people are all important, we at Fujitsu cannot do this alone. We will work on this issue with the spirit of co-creation for success."
CEO, 1QB Information Technologies Inc.
Professor, Department of Electrical and Computer Engineering, University of Toronto
Vice Head of the Digital Services Business, Corporate Executive Officer, Fujitsu Limited
Director and Head of the Digital Annealer Project, Fujitsu Laboratories Ltd.