Digital Annealer Introduction - fujitsu.com · Annealing Method Classical Approach Metal is heated...

Copyright 2019 FUJITSU LIMITED

Digital Annealer

Introduction

December 21, 2018

FUJITSU Quantum-Inspired Computing Digital Annealer

Patent pending including related technology

Fujitsu Limited

Quantum Computing is one promising prospect as a next generation computer


1990 2010 20202000

Data Volume

103

102

1

101

104

10nm

20nm40nm

180nm

250nm

90nm

Today

GAP

UnstructuredData

StructuredData

350nm

2025

2002

Moore’s Law foretells of improved CPU density and performance

Performance improvement hasslowed due to power constraints

The limit ofMoore’s Law

Power efficiency performance of general CPU “Relative value”

Reaching the Limit of Moore’s LawComputers must process increasingly massive and complex data at higher and higherspeeds in order to support digital transformation in society and business. Moore’s Law* is approaching its limit, threatening the drastic compute performance required in the coming future.

*Moore's Law: An empirical rule in the semiconductor industry stating that the number of transistors in a dense integrated circuit doubles every 18~24 months.

Power Efficiency

Performance

Year

1


Digital AnnealerA new architecture that solves "combinatorial optimization problems" at high speed with digital circuits inspired by quantum phenomena

Stable operation with digital circuit, and easy miniaturization

Easy mapping of more complex problems with a fully-connected architecture

Digital Annealer

Difficult to maintain a quantum state

Limits in connection and expansion

Quantum ComputersEasy to apply to actual problems

Still in the research stage ...

2


Digital Annealer Positioning

Quantum Computing Technology

DevelopmentOrganization

Method

Stable operation atroom temperature

Available in traditionalrack / server room

Requires cryogenic environment, but still technically unstableLarge cooling device required

Quantum Architecture

FujitsuDigital Annealer

D-Wave systemsD-Wave 2000Q

NECHitachi, Ltd.

CMOS AnnealingMachine

Non-Quantum Architecture

Google Microsoft IntelIBM

IBM Q

Quantum Gate SystemComputation using quantum gates and

quantum bits that can be in a superposition of 0 and 1 states

量子イジングマシン方式

Superconducting Circuit

Ising Machine SystemMap a problem to a mathematical model of interacting magnets, then search for solutions

Specialized for combinatorial optimization problems

Annealing

Digital Circuit

LaserNetwork System

National Institute of Informatics, NTT and others

Japan Cabinet Office ImPACT"Quantum Neural Network"

Optical Parametric Oscillation

Quantum Annealing

Analyzed by Fujitsu

Unlike quantum computers, Digital Annealer does not require an extremely low temperature environment. Digital Annealer operates stably at room temperature.

Digital Annealer makes use of the annealing method, specialized for combinatorial optimization.

3

Annealing Process in Metallurgy


What is Quantum Annealing? The algorithm’s name comes from the annealing process used in metallurgy

Blocks are placed randomly, then the entire system is“shaken”. The shaking is gradually reduced, and the shapes quickly fit together.

Blocks are placed in sequence. Process restarts if a solution is not found. Repeated until a solution is found.

Annealing Method Classical Approach

Metal is heated to a high temperature, and the structure stabilizes as it is slowly cooled(=low energy)

The most stable state has minimum energy The minimum value

4

Choose the number of cities Visit each city only once

Evaluate based on travelling distanceShortest route


Combinatorial Optimization ProblemsSeek combinations or sequences that satisfy given constraints, with the goal of finding

the best out of all available combinations

City CityCity

City

City CityCity

City

City CityCity

City

With 5 cities 120 possible routes. With 32 cities 2.63x1035 possible routes

Number of cities: N Constraint Optimal Solution

“Which is the shortest route that visits each city exactly once and returns to the origin city?”

Example: The Traveling Salesman Problem

The number of combinations increases exponentially!5

The resulting bits show the shortest path solution

City B ⇒ City A ⇒ City D ⇒ City C ⇒ City E

Create an expression to define the interaction between bits (distance

between cities)

Use Digital Annealer to Solve Combinatorial Optimization Problems

“Which is the shortest route that visits each city exactly once and returns to the origin city?”

INPUT OUTPUT

Formulation

City A City B City C City D City ECity A City B City C City D City E

0th

1st

2nd

3rd

4th

Cities to Visit

0th

1st

2nd

3rd

4th

0

1

0

0

0

1

0

0

0

0

0

0

0

1

0

0

0

1

0

0

0

0

0

0

1

Cities to Visit

Each represents one bit

Order to Visit

The Traveling Salesman Problem


3. Execute on Digital Annealer2. Build Formula1. Capture Problem in Ising Model

6

Applicable to New Areas

Improve Precision and Reduce Time to solve existing combinatorial

optimization problems


Combinatorial Optimization Problems Across All Industries & Business

Advanced Healthcare

New Material Development

Distribution RoutePlanning

UtilitiesManagement

Demand Forecasting

Materials & Compounds

Network ConfigManagement

LogisticsHospitality

Management

DrugDiscovery

Portfolio Management

DigitalMarketing

Applicable Area Examples

AutonomousVehicles

7


Problem and Application Examples Well-Suited for Digital Annealer

Optimization Problems

Maximum IndependentSet Problem

Minimum Set Cover Problem

Candidate Overlay Problem

Max Cut Problem

1. Molecular Similarity Search

2. Big Data Visualization Toolkit

3. Traffic Route Optimization

4. Financial Portfolio Optimization (QHRP Method*1)

5. Shelf Location Optimization

Molecular similarity identificationin compounds

Visualizing and clustering big data

Optimize route selection while minimizing route overlap to ease congestion

Select investment portfolio assetsbased on low correlativity

Optimize shelf placement to improve factory parts pickup

*1 QHRP: Quantum-inspired Hierarchical Risk Parity

8


1. High Precision Molecular Similarity Search for Drug Discovery

The conventional Finger Print method determines the presence or absence of an atomic group, but does not consider the molecular shape. Thus, a precise search cannot be performed.

By converting the molecular structure to a graph and handling atomic groups as nodes and bonds as edges:• Precision is improving by considering molecular shape• Calculations are performed at high speed by Digital Annealer

• Highly precise molecular similarity search becomes possible• Expected to improve the efficiency of drug development leading to new

highly effective medicines

Issues

Technique

Results

Use Case

Combinations of atomic groups results in a huge volume of calculations

Improve efficiency with Digital Annealer

Conventional Method

Digital Annealer Benefit

Finger Print

Matching without consideration of molecular shape

Graph Theory

Highly accurate by matching molecular

shapes

Chemical & Pharmaceutical

Contributing to the development of highly effective medicines

Finger Print: A method of representing the presence or absence of an atomic group as 0 or 1 and expressing the molecule as a Boolean vector

9


2. High-Speed Clustering for Big Data Utilization

As the importance and prevalence of big data increases, high-speed data processing is necessary to effectively derive business insights

High precision clustering with hierarchical structures is implemented by compressing high dimensional data and segmenting it into portions that can be clustered

• Clustering is accelerated from several hours with conventional methods to just a few minutes with Digital Annealer

• Large scale data sets can be visualized and analyzed• The level of clustering can be changed to enhance analysis

Issues

Technique

Results

Use Case

Marketing

Visualizing large-scale datasets for more accurate analysis

Conventional Digital Annealer

Customer data for 40,000 people Grouped by similar data

Highly accurate clustering with Digital Annealer

Changeable according to the level of clustering

Coarse grained

Fine grained

10


3. Route Optimization to Reduce Traffic Congestion

Total Travel time

Shortest Route

With Digital Annealer

With conventional routing systems, there is a tendency to assign the shortest distance route, leading to traffic congestion in the city center

• Optimize route selection to avoid overlap• Prioritize route options by adding conditions, such as: speed

limits, number of lanes, etc.

• Reduce traffic congestion by up to 40% by dispersing traffic• Apply to cases of iterative simulation used for road development planning• Applicable to other routing problems, such as warehouse collection and

distribution, AGV (Automated Guided Vehicles), and network traffic 0

50

100

150

DigitalAnnealer

58hrs

Total Travel Time

101hours

58hours

Issues

Technique

Results

Use Case

LogisticsReduce overall travel time by distributing routes throughout a city

or factory to avoid congestion

Conventional Digital Annealer

Select Shortest Path

Optimized Route Selection

Degree of Congestion

High

Low

Example Max

Speed (km/h)

10

20

40

50

60

Degree of Congestion

High

Low

Example Max

Speed (km/h)

10

20

40

50

60

Shortest Route

Method101hrs

40% Decrease

Immediately recalculate routesaccording to traffic condition changes

11


4. Investment Portfolio Optimization Through Risk Diversification

The commonly used Minimum Variance (MV) method for portfolio optimization is susceptible to the influence of market fluctuation

Quantum-inspired Hierarchical Risk Parity (QHRP) portfolio optimization method provides for:• The clustering of assets into a tree diagram based on risk

correlations • Composition of risk-diversified portfolios with low correlativity

• Create portfolios with resistance to market fluctuations that continue to provide stable returns

• 60% higher Sharpe Ratio compared to MV method

Issues

Technique

Results

Use Case

Finance

Instant clustering for the correlation of 500 stocks to compose a risk-resistant portfolio

Risk/Return Ratio

Results (Sharpe ratio comparison)

Digital Annealer Technique

Clustering Assets from CorrelationsAsset Price Change Variance Matrix

Company

ACompany

BCompany

C …

Company A 1 0.23 0.85

Company B 0.23 1 0.64

Company C 0.85 0.64 1

…

…

Express all correlations among assets(price changes between two assets) using

variance-covariance matrices Companies: A C D B

Lehman Shock

QHRP

MV

S&P500 Investment Results (2005 – 2011)

12


5. Factory Parts Pick Up Optimization

• High-mix, low-volume factory production requires a large variety of parts for each product. Time and labor required dependent on the experience level of each worker

• Inconsistent and inefficient parts pick up process

• Routes and shelf population are minimized as combinatorial optimization problems

• Correlation of frequently used shelves identified

• Even inexperienced workers can realize efficient parts picking• Travel distances reduced by up to 45% per month through route and

shelf location optimization• Optimization methods to be deployed to other factories, as well as

other processes such as warehouse management

Issues

Technique

Results

Use Case

Manufacturing

Reduce travel distance for warehouse parts pick up by up to 45%Now in use at Fujitsu IT Products

Very complicated parts picking routes required experienced workers

Digital Annealer provides optimum picking routes displayed on a tablet

Warehouse area: 1000m²Number of parts: 3000

13


5. Case Study:

Factory Parts Pick Up Optimization - At Fujitsu IT Products

Case Study: http://www.fujitsu.com/global/digitalannealer/case-studies/201804-fjit/

Company Profile

Digital Annealer Project Schedule

Manufacturing

Company Name

Location

Capital

Establishment

Industry

Employees

Fujitsu IT Products Limited

1-1, Kasajima-to, Kahoku-shi, Ishikawa, 929-1196, Japan

100 million yen (wholly owned subsidiary of Fujitsu Limited)

April 1, 2002 455 people

Manufacture of servers, supercomputers, storage systems , software, etc.

Servicein February 2018

PoC fromOctober 2017

Discussion fromSeptember 2017

Source: http://www.fujitsu.com/jp/group/fjit/ (Japanese)

14

http://www.fujitsu.com/global/digitalannealer/case-studies/201804-fjit/

http://www.fujitsu.com/jp/group/fjit/


5. Case Study:

Factory Parts Pick Up Optimization - ConfigurationManufacturing

Before

With Digital Annealer

Tablet

Request a formula and distance between shelves from Digital Annealer via REST-API

ManagementServer

Optimum route provided

a b c d e f

a 0 1 2 3 6 7

b 1 0 1 4 5 6

c 2 1 0 5 4 5

d 3 4 5 0 4 4

e 6 5 4 4 0 1

f 7 6 5 4 1 0

①

②

③

④

⑤ ⑥

Shelf positionDistance

between shelves

Part1: Shelf aPart3: Shelf cPart6: Shelf f

Shelf ac: 2Shelf cf: 5Shelf af: 7

Shelf Position DB

Output Optimal route

Pick up done based on the best route

Shelf positions and distances between shelves are put in the Shelf Position DB

1

2

3

4

5

Tablet

Shelf a → c → fPart 1 → 3 → 6

Shelf c→f→aPart 3→6→1

Shelf ac: 2Shelf cf: 5Shelf af: 7

＋ Mathematical Expression

Newly Implemented

Digital AnnealerCloud Service

0

1

1

1

0

0

1

1

Digital Annealer

Shelf Position DBComponent Parts DB

Part1: Shelf aPart3: Shelf cPart6: Shelf f

Part1: Shelf a Part2: Shelf bPart3: Shelf c Part4: Shelf dPart5: Shelf e Part6: Shelf f

Product A-Part 1-Part 3-Part 6

Order Input

Product A

Sales Factory

Goal: Leverage existing operation, and create a new shelf position databaseto use Digital Annealer to calculate optimal pick up routes

Based on information displayed on the tablet, parts picked up

15


Easy to use with total bit couplingConnectivity

High precision with 64bit graduationsPrecision

Stable operation at roomtemperature with digital circuits

Stability

Applicable to real world problems with the stability and

balance of scale, connectivity, and precision

Digital Annealer Advantages

Ready to scale up for 8,192bit problemsScale

16

Scale Connectivity Precision


What Are Scale, Connectivity and Precision?

The Traveling Salesman Problem

City CityCity

CityUnder the constraint that each city must be visited only once, find the shortest route (minimum distance)

Scale:Number of cities to visit

that can be handled

Connectivity:All distances betweencities can be defined

Gradations:Accuracy of distance

between cities

8192 bits Total Coupling 64bit Gradations

0th

1st

2nd

3rd

4th

City A B C D E

0th

1st

2nd

3rd

4th

VisitOrder

Cities to Visit

0th

1st

2nd

3rd

4th

Near Medium Far

VisitOrder

VisitOrder

For example, 3 gradations can be expressed as near, medium and far.

Cities to VisitCities to Visit

City A B C D E City A B C D E

17


Digital Annealer Features

Differe

ntia

tors

Higher escape probability when local solution is detected

StateOptimal Solution

Local Solution

Offset

Escape!B

C

A

0

1

1

1

0

0

1

0

Bit 1

Bit 2

…

Bit i

…

Total Coupling

• Increased escape probability from local minimum energy states using hardware offsetting

• Represent large-scale problems of up to 8192 bits

• High precision with full coupling and 264 bit gradations

• Easy to use with fully connected bits

• Next bit inversion found using parallel search to increase processing speed

• Stochastic parallelism provides significant acceleration

Current set of bits

Parallel Search

Set bits aftersingle bit updated

・・・

Trial1 Trial 2 Trial N

Evaluation

Bit 1Inversion

Bit 2Inversion

Bit NInversion

Evaluation Evaluation

Eval

uati

on V

alue

Ready for Many Business Tasks

High Speed Processing withLocal Solution Escape and Parallel Search

18


Specification ComparisonApplicable to real world problems with the stability and balance of

scale, connectivity, and precision

Digital Annealer(2nd generation)

Company A Company B

ImplementationTechnology

Digital Circuit(Using existing technology)

Superconductive Circuit(Cryogenic cooling required)

Non-Linear Optical(1km ring device)

Number of Bits 8192 2048 2048

Amount of Coupling Total Coupling6 - Partly Coupled(64bit total coupling

equivalent)

Total Coupling

EvaluationAccuracy 64bit Gradations 32 Gradations 3 Gradations

19


Digital Annealer

Services Overview


Combinatorial Optimization Problems Faced in Actual Business


Digital Annealer Service OverviewCloud Service and Technical Service launched in May 2018

TechnicalService

CloudService

On-Premises(planned)

FUJITSU Quantum-Inspired Computing Digital Annealer Cloud Service Digital Annealer Technical Service

Realize high speed processing forcombinatorial optimization problems

Customer Support forDigital Annealer utilization

21

Digital Annealer Cloud ServiceFUJITSU Quantum-Inspired Computing Digital Annealer

Ising model

Mathematical expressions

Variables

Constraint conditions


Customer

WebAPI

WebAPI

Formulation

0

1

1

1

0

0

1

1

Find Optimal Solutions

Request

Response

Netw

ork

Via Web API, submit Ising model mathematical expressions and constraints,

then receive solutions in seconds

Copyright 2019 FUJITSU LIMITED22


*It may take time to start calculating the mathematical model.

Help Desk Service is mandatory for all Service Levels

Cloud Service MenuFUJITSU Quantum-Inspired Computing Digital Annealer

1st generation: 1024bit 2nd generation: 8192bit

Service Level Gen Fee Category Web API Type Remarks

Premium 1st Fixed monthly Dedicated SynchronousPlease contact Fujitsu in advance

to schedule environment deployment

Standard

1st Basic fee (fixed monthly)+metered rate (by usage)

Shared*

Synchronous

Asynchronousoptional

2nd

Selectable

Basic fee (fixed monthly)＋metered rate (by usage)

Fixed monthly

Trial

1st Basic fee (fixed monthly)＋metered rate (by usage)

Shared*

Synchronous


Limited to 6 months2nd

Selectable

Basic fee (fixed monthly)＋metered rate (by usage)

Fixed monthly

Academic1st

Fixed monthly Shared*

Synchronous


Limited to universities andgraduate schools for educational

and research purposes2nd

23

The API requests calculation results synchronously.The API returns when the calculation process has completed.

A calculation request is issued and the API returns. A separate request is issued to obtain the calculation result.

Cloud Service Web APIs FUJITSU Quantum-Inspired Computing Digital Annealer

Synchronous Web API (basic) and Asynchronous Web API (optional) offerings:

Multiple requests can be made for large problemsThe problem is processed in real time

Synchronous Web API Asynchronous Web API

Ising Model Mathematical

Expressions Variables Constraint

Conditions


CustomerRequest

OptimalSolution

…

Ising Model Mathematical

Expressions Variables Constraint

Conditions

CustomerRequest

0

1

1

1

0

0

1

1


0

1

1

1

0

0

1

1…

Netw

ork

Netw

ork

OptimalSolution

Response(with Acceptance ID)

Request for result corresponding to ID



Help Desk ServiceFUJITSU Quantum-Inspired Computing Digital Annealer

Service Details for Digital Annealer Cloud Service Subscribers:

Weekdays 9:00-17:00 (JST) *Closed on Japanese holidays

Inquiries received outside service hours will be processed after 9:00am on the following business day

The following inquiries are excluded from the Help Desk Service:・Processing speed tuning (performance evaluation resulting from customer’s APP design, implementation and operation, etc.)・Consulting (advice on creation, design, implementation and operation of mathematical models)・Disclosure of information related to our cloud service environment and logs・Calculation result accuracy

Service Hours

• Questions about specifications, settings and usage of the Digital Annealer Cloud Service• Questions, investigation, workarounds when the Digital Annealer Cloud Service does not function correctlyAcceptable Questions

EmailInquiry Contact Method JapaneseLanguage

Important Notes

Calculated as 5% of the Web-API usage service fee Service Fee

Trouble Notification Maintenance Notification Service Detail / Update Guide

Detail

Method

Notification from the time the issue occurs, functions impacted, etc.

Portal/Email

Detail

Method

Detail

Method

Notification of Digital Annealer Cloud Service planned or emergency maintenance date/time, functions impacted, etc.

Portal/Email

Notification of new functionality and improvements made to the Digital Annealer Cloud Service

Portal/Email

25

Digital Annealer Technical Service

Complete support from introduction to operation, leveraging Digital Annealer for customer business

Digital Annealer Technical Service - OutlineDigital Annealer Technical Service

Technical experts with advanced mathematics knowledge and data analysis capabilities support you in resolving your optimization problems

Training for EngineersTraining

Tuning Support ServiceTuningSupport

Convert customer issues into formula

Apply formulation to customer issues

Applicationimplementation

Maintenance,Support

Assessment Deployment Support Implementation Operation


Operation PhaseImplementation Phase

26


Digital Annealer Technical Service DetailsDigital Annealer Technical Service

Verify the customer’s problem can be converted to a combinatorial optimization problem, and generate a mathematical model to solve with Digital Annealer.

PoC planning for customer Digital Annealer use. Evaluate mathematical model implementation and results. Support requirements definition for customer Digital Annealer deployment.

Develop and deploy application for use with Digital Annealer: Deploy on customer systems, establish connection to Digital Annealer, process input / output data, and perform post processing of output.

Formulation Verification Development Support Implementation

Answer questions and solve problems for the deployed Digital Annealer application.

F2F instructor-led Digital Annealer training:• Basic: Digital Annealer features and

mechanisms•Hands-on: Solve optimization

problems using Digital Annealer Cloud Service

Problem solving support including parameter tuning for Digital Annealer Cloud Service customers.

Operation Training Tuning Support

27

Combinatorial Optimization Problems Faced in the Real World


Complete Solutions for Real-World Business Problems

Issues Solutions

Ising ModelMathematical Formula

Find OptimalSolution

OptimizationCalculation

BusinessApplication

Technical ServiceCloud ServiceTechnical ServiceFormulation

Analysis/Verification

28


Digital Annealer Project Example

Typical Project Duration: 4-6 months

• Procedure Agreement

• Conclude NDA (As needed)

• Pre-meeting(2-3 times)

• Technical Introduction• Technical Verification

• Select target business• Agreement on Expectations

• PoC Plan Agreement

• Formulation• Ising Model Construction

• PoC on Digital Annealer-Confirm Functions-Performance, UX, etc

• PoC Result Summary

Available Digital

Annealer Services

2~3 months

NextStepPoCModel ConstructionSelect business

target for PoCTechnical Check

Pre-PhasePhase 1 Phase 2 Phase 3 Phase 4

2~3 months

Trial: Maximum 6 months (Fee-based)

Technical Service (Fee based)

Premium/Standard (Fee based)

29

A new service in thequantum computing area

Solving combinatorial optimization problems to control Automated

Guided Vehicles (AGVs)

Touhoku University

New marketing technology research and development for real-world business problems

RecruitCommunications

Co., Ltd.

Technical verification of portfolio optimization in asset

management

Mitsubishi UFJ TrustInvestment Technology

Institute Co., Ltd.

Technical verification of production lines for high-mix

low-volume models in factories

Fujifilm Corporation

Fixstars Corporation


Digital Annealer Case Studies

Participation inthe MITOU Program

of METI / IPA

Using Digital Annealer to foster talent and start-up companies

in technical fields

30


Digital Annealer

Roadmap


Digital Annealer Roadmap

Scale: 1024 bit

Digital Annealing Unit

1st Gen Cloud Cloud2nd Gen

2018

Technical Service

May On-Premises Service 4QDec

2019

Expand applications from technical verification to real-world business value

Next Generation

Precision: 16 bit 65536 Gradations

Max Scale: 8192 bit

Max Precision: 64 bit1.845x1019 Gradations

Large-Scale Parallel Processing

32

Size (bit)

Precision (bit)

Company A

Company B

Company C

2nd Generation

Priority for Size

1st Generation


Priority for precision and priority for size

can be switched

16bit

64bit

Application Areas Surpassing the Competition - 2nd Generation

8192bit1024

bit

33

Accuracy

2nd generation

Priority for size

1st generation

1024bit

16bit

8192bit

64bit

Size

Precision

2nd Generation

Priority for Size

1st Generation

1024bit

16bit

1,000,000 bit

Supports up to 1,000,000 bitswith software partition technology & multichip support


Application Areas Surpassing the Competition – Next Generation

Develop further large-scale application technology2019

8192bit

64bit

34


Tackling Extremely Large Scale Problems

Hardware technology links multiple Digital Annealer Units (DAUs) to execute large scale parallel processing

DAU

Depending on the characteristics of the problem, multiple splitting

methods available

Find the optimal solutionusing a solution search flow

Large scale problems split into manageable portions by software

DAU DAU DAU

Fujitsu is developing a problem-splitting method to solve extremely large-scale problemsThis method extracts a portion of a problem according to problem characteristics; processing each portion on Digital Annealer, then assembling the output from multiple calculations to derive the total optimal solution using search flow.

This method allows a single 2nd generation (8192 bit) Digital Annealer to solve 100,000 bit scale problems

Further software and hardware enhancements will expand the scale to 1 million bits (planned for CY2019)

35

Drug candidate search can be drastically shortened from 6 months to just a few days,accelerating the development of middle molecule medicine


Accelerate the Development of Middle Molecule Medicine

Search for the most stable structure binding relationship within amino acids, when each amino acid is modeled and placed on a lattice point

Linear molecular structureof amino acids

Represents one amino acid

Search for optimal structure

Return to the molecularstructure

Amino acidtype

Apply Digital Annealer + the problem-splitting method for simulation to find stable structure middle molecule drug candidates jointly with ProteinQure Inc.

Effectiveness demonstrated with 30K bit scale problem

36

Digital Annealer Global Rollout

Japan

• Cloud Service • Technical Service

launched May2018

EMEIA

• Cloud Service (Planned)

• Technical ServiceCustomized Support

APAC &Oceania




Digital Annealer Cloud Service rolling out in each region starting from Japan

Technical Service is under development globally Americas



37

New AI Headquarters (AI HQ) - Established October 2018

Creating a core base in Vancouver to roll out Fujitsu's AI business and accelerate the application of AI to customers worldwide using Digital Annealer as a key technology

Americas

AIHQ

APAC

EMEIA Japan

Deployment of advanced global use cases

• Distribution of technology and solutions

• Expert technical support

Leading an AI ecosystem

• Expansion of platform users

Creation of AI core business

• Planning, execution, and roll out • Company-wide AI strategies

Oceania



Digital Annealer

Partnerships



Expanding Digital Annealer Applications Through Partnerships

1QBit middleware implementedon Digital Annealer

Collaborative researchin cutting-edge areas

Confronting new issues in society

Digital Annealer

Promotion of combinatorial optimization as a way to

solve societal issuesExpansion of application areas

40

Digital Annealer


Partnership with 1QBit

HardwareThe processing power

to solve problemswith high speed and

high precision

MiddlewareMathematical formulas

and algorithms for computation

The world’s #1 vendor of software for quantum computers

Conducting joint business around the world

Planning to incorporate Digital Annealer in 1QBit Cloud Service in CY2019

41


Partnership with University of Toronto

Smart Traffic

Finance

Network

Medical

World-class research university in the fields of AI and quantum computing

New joint research center established with the University of Toronto in November, 2017:

Fujitsu Co-Creation Research Laboratory at the University of Toronto

Joint research in smart transportation, networks, finance, and healthcare fields

42


Joint Research with the University of Toronto- Cancer Radiation Therapy

Huge number of irradiation patterns (number of combinations) with variations such as range,

direction, and intensity of irradiation

Even when the beams are from only one direction, the number ofcombinations would be: 10150

In case of irradiation against a 1cm2 tumor with a precision of 1mm2 and 32 intensity levels from one direction

Huge computational load required for simulation before a treatment plan can be made.

Current technology needs multiple hours to a few days to calculate the combinatorial optimum.

Digital Annealer takes only a few minutes

Cancer

Vital Organs

Intensity Modulated Radiation Therapy (IMRT)

43


Partnership with Waseda University

Candidate Topics

Traffic/Logistics Delivery Optimization

DigitalMarketing Data Analytics

FinanceStock Price

Fluctuation Prediction

Comprehensive Collaborative Activity Agreement Established

Solve Real-World Societal Issues

Healthcare Disaster Remediation

Chemical, Pharma Traffic/

Logistics

ManufacturingEnergy

SecurityDigital

MarketingFinance

Private Company

Joint research into solving combinatorial optimization problems with Digital Annealer

ResearchTopics

Starting April 2019: Collaborative research using Digital Annealer to solve real-world combinatorial optimization problems

Drawing research topics from Waseda University’s entire body of research

Joint research site opened in March 2018: Fujitsu Co-Creation Research Laboratory at Waseda University

44


Digital Annealer Reference Sites & Press Releases

Digital Annealer Websites:

Japanese English

http://www.fujitsu.com/global/digitalannealer/

http://www.fujitsu.com/jp/digitalannealer/

Press Releases

Fujitsu Laboratories and Waseda University Agree to Comprehensively Collaborate on Digital Annealer Research

September 19, 2018Fujitsu Technology to Solve Combinatorial Optimization Problems for Medium-Sized Drug Discovery

September 18, 2018Fujitsu Quantum-Inspired Digital Annealer Cloud Service to Rapidly Resolve Combinatorial Optimization Problems

May 15, 2018

Fujitsu Initiates Joint Research with Recruit Communications on Marketing Technologies Using "Digital Annealer"

January 29, 2018

Exhibited at "CEATEC JAPAN 2017“

September 27, 2017Fujitsu Technology Facilitates Application of Combinatorial Optimization Methods to Real-World Problems

September 20, 2017

Fujitsu Laboratories and University of Toronto Enter Strategic Partnership

September 20, 2017Fujitsu and 1QBit Collaborate on Quantum Inspired AI Cloud Service

May 16, 2017Fujitsu Laboratories Develops New Architecture that Rivals Quantum Computers in Utility

October 20, 2016

http://www.fujitsu.com/global/about/resources/news/press-releases/

YouTube

Digital Annealer public channel

http://www.youtube.com/channel/UCo0c9YwYOHXLwJnNA_moEJQ

Fujitsu and TC3 Promote Quantum Inspired Digital Annealer Next-Generation Architecture in Topcoder Contest

January 15, 2019Fujitsu Launches Next Generation Quantum-Inspired Digital Annealer Service

December 21, 2018Fujitsu Drives Quantum-Inspired Project to Help Solve NatWest’s Complex Optimization Challenges

October 2, 2018

45

Digital Annealer Introduction - fujitsu.com · Annealing Method Classical Approach Metal is heated...

Documents

Transcript of Digital Annealer Introduction - fujitsu.com · Annealing Method Classical Approach Metal is heated...