Fujitsu Laboratories Advanced Technology Symposium 2018 October 9, 2018
Trust and Co-creation in the Digital Era Shigeru Sasaki Fujitsu Laboratories Ltd. CEO Fujitsu Limited CTO 2
FLATS 2017 Quantum Computing: Leading the Next Generation IT Revolution October 11, 2017 Computer History Museum 3
FLATS 2017 Quantum Computing: Leading the Next Generation IT Revolution October 11, 2017 Computer History Museum 3
FLATS 2018 Make AI Trustworthy! Explainable and Ethical AI for Everyone 4
FLATS 2018 Make AI Trustworthy! Explainable and Ethical AI for Everyone 4
FLATS 2018 Make AI Trustworthy! Explainable and Ethical AI for Everyone 4
Fujitsu s R&D Strategy 5
Fujitsu as a Technology Company To resolve societal issues and bring happiness to the people around the world 6
Fujitsu as a Technology Company To resolve societal issues and bring happiness to the people around the world 6
Announced in 1976 Reliability and Creativity 7
Announced in 1976 Reliability and Creativity 7
Changes in the Digital Era Interdependency among company, people, business, data, and system is extremely high Complexity Immensity" Difficulty in judging the Quality" Noise contamination inevitable "Trusted World accelerates "Co-creation" 8
Changes in the Digital Era Interdependency among company, people, business, data, and system is extremely high Complexity Immensity" Difficulty in judging the Quality" Noise contamination inevitable "Trusted World accelerates "Co-creation" 8
Trust and Co-creation in the Digital Era Reliability Creativity Trust Co-creation 9
R&D Strategy Generating innovations through Digital Trust and Co-creation and promoting societal and economic development Trust : Create Trust in society and customers business environment, by leveraging cutting-edge technologies as well as our expertise Co-creation : Enhance and expand the customers core business for mutual growth through cutting-edge technologies 10
R&D Strategy Generating innovations through Digital Trust and Co-creation and promoting societal and economic development Trust : Create Trust in society and customers business environment, by leveraging cutting-edge technologies as well as our expertise Co-creation : Enhance and expand the customers core business for mutual growth through cutting-edge technologies 10
The world that Trust connects in the Digital Era Governance Personal Authentication Cross-Industry Business Business Data Traceability Management Compliance Business Contract Social Responsibility People System Transparency Data People Risk Management GDPR Trust Cyber Security Rule/ Regulation Data System Information Ethics and Morality Anonymization / Concealing Explainability Robustness / Availability Accountability 11
The world that Trust connects in the Digital Era Governance Personal Authentication Cross-Industry Business Business Data Traceability Management Compliance Business Contract Social Responsibility People System Transparency Data People Risk Management GDPR Trust Cyber Security Rule/ Regulation Data System Information Ethics and Morality Anonymization / Concealing Explainability Robustness / Availability Accountability 11
The world that Trust connects in the Digital Era Governance Personal Authentication Cross-Industry Business Business Data Traceability Management Compliance Business Contract Social Responsibility People System Transparency Data People Risk Management GDPR Trust Cyber Security Rule/ Regulation Data System Information Ethics and Morality Anonymization / Concealing Explainability Robustness / Availability Accountability 11
Trust can be easily damaged Trust even by slightest misconduct 12
Fujitsu s AI Strategies 13
Achievement of Trust and Co-creation by Cutting-Edge Technologies Cross-Industry Compliance Business Governance Contract Data Innovative Reliable AI Business Business Management Computing Personal Solving that Develops System that Social Responsibility Social Authentication Issues the Society Facilitates the Value Information Data Traceability Creation Cycle and Communication Management System People Management System Data Engineering People in the 5G Era for Transparency Supporting Connecting Cyberspace Digital Co-creation Risk Management Digital Security and Physical World Sensecomputing GDPR Cyber Security Rule/ Regulation Systems for Human-Machine New Materials/ Trust Protecting Collaboration Devices that Society from Data through System Revolutionize Human Emotion Information Potential Threats the Existing Paradigm Explainability Understanding Ethics and Morality Accountability Anonymization / Concealing Robustness / Availability 14
Fujitsu Laboratories AI Strategy Knowledge Sharing & "Quick Decision Making Explainable AI Knowledge Graph + Deep Tensor Wide Learning Human-in-the-Loop Discovery Science Solving the Unsolvable Issues Digital Annealer Domain Specific Computing 15
Fujitsu Laboratories AI Strategy Knowledge Sharing & "Quick Decision Making Explainable AI Knowledge Graph + Deep Tensor Wide Learning Human-in-the-Loop Discovery Science Solving the Unsolvable Issues Digital Annealer Domain Specific Computing 15
Explainable AI Reliable, Understandable and Controllable to Human Deep Tensor + Knowledge Graph: To explain Reason and Basis Wide Learning: Support of highly accurate and transparent judgement by learning from little training data Deep Tenor + Knowledge Graph Input Deep Tensor Extended error Backpropagation Classification A Classification error Classification B Auto generation of feature amount Output Output: Estimation and Reason Findings Inference factors Wide Learning Knowledge Chunk : Simple models that human understands OUTPUT g a b c d e f Knowledge Graph Consisting Basis by Knowledge Graph INPUT 16
Explainable AI Reliable, Understandable and Controllable to Human Deep Tensor + Knowledge Graph: To explain Reason and Basis Wide Learning: Support of highly accurate and transparent judgement by learning from little training data Deep Tenor + Knowledge Graph Input Deep Tensor Extended error Backpropagation Classification A Classification error Classification B Auto generation of feature amount Output Output: Estimation and Reason Findings Inference factors Wide Learning Knowledge Chunk : Simple models that human understands OUTPUT g a b c d e f Knowledge Graph Consisting Basis by Knowledge Graph INPUT 16
Explainable AI Reliable, Understandable and Controllable to Human Deep Tensor + Knowledge Graph: To explain Reason and Basis Wide Learning: Support of highly accurate and transparent judgement by learning from little training data Deep Tenor + Knowledge Graph Input Deep Tensor Extended error Backpropagation Classification A Classification error Classification B Auto generation of feature amount Output Output: Estimation and Reason Findings Inference factors Wide Learning Knowledge Chunk : Simple models that human understands OUTPUT g a b c d e f Knowledge Graph Consisting Basis by Knowledge Graph INPUT 16
Digital Annealer The problem that cannot be solved yet is solved Solving large-scale combinatorial optimization problems with high speed and accuracy Support quick decision making and judgement of human 1 1 Technologies 0 Applications Cobalt source Radiation beam 0 0 D A Y A Q W L K Diseased part Collimator 0 1 1 Inter-bit Full Connection Quantum-Inspired Computing A D Q K Y W A L Combinational Optimization Drug Discovery Source: National cancer research center https://ganjoho.jp/public/dia_tre/treatment/ radiotherapy/rt_03.html Advanced Cancer Treatment 17
Digital Annealer The problem that cannot be solved yet is solved Solving large-scale combinatorial optimization problems with high speed and accuracy Support quick decision making and judgement of human 1 1 Technologies 0 Applications Cobalt source Radiation beam 0 0 D A Y A Q W L K Diseased part Collimator 0 1 1 Inter-bit Full Connection Quantum-Inspired Computing A D Q K Y W A L Combinational Optimization Drug Discovery Source: National cancer research center https://ganjoho.jp/public/dia_tre/treatment/ radiotherapy/rt_03.html Advanced Cancer Treatment 17
Digital Annealer The problem that cannot be solved yet is solved Solving large-scale combinatorial optimization problems with high speed and accuracy Support quick decision making and judgement of human 1 1 Technologies 0 Applications Cobalt source Radiation beam 0 0 D A Y A Q W L K Diseased part Collimator 0 1 1 Inter-bit Full Connection Quantum-Inspired Computing A D Q K Y W A L Combinational Optimization Drug Discovery Source: National cancer research center https://ganjoho.jp/public/dia_tre/treatment/ radiotherapy/rt_03.html Advanced Cancer Treatment 17
Human-in-the-Loop Human s Trust: Enhance abilities of judgment and decision making of human which uses technologies AI and System s Trust: Minimize human errors Social Systems with ICT Decision making / Judgment 18
Human-in-the-Loop Human s Trust: Enhance abilities of judgment and decision making of human which uses technologies AI and System s Trust: Minimize human errors Social Systems with ICT Decision making / Judgment 18
High Societal Acceptability of AI Strategy to develop AI market surely Societal Acceptability Enhancement by industryuniversity collaboration The world where society and AI can coexist smoothly Societal acceptability that advances societal implementation of AI Privacy and Security ELSI (Ethical, Legal and Societal Issues) Protection of intellectual property rights (method / data / Model)) 19
High Societal Acceptability of AI Strategy to develop AI market surely Societal Acceptability Enhancement by industryuniversity collaboration The world where society and AI can coexist smoothly Societal acceptability that advances societal implementation of AI Privacy and Security ELSI (Ethical, Legal and Societal Issues) Protection of intellectual property rights (method / data / Model)) 19
Practical Use Cases: AI Societal Implementation 20
Inspection of Windmill Turbine Blade Siemens Gamesa Wind Power Inspection time: Reduction in 1.25 hours Efficiency improvement of 83%, Reduction for 34,000 hours a year FIAIR as AI package of Non-destruction inspection was released in Europe from September, 2017 Size and Number Length: 75m Speed: 100 km/h Issues Inspection time for 1 blade Non-destruction Test by Ultrasonic Scanning from 30 to 45 min. Inspection from Ultrasonic Image Highly-skilled inspectors check 1 blade from 6 to 8 hours Every year 5,000 Blades Inspection End Inspection from Ultrasonic Image Detection of Defects by Deep Learning Inspection time of Ultrasonic images Reduction in 1.25 hour 21 37 Technologies Manual inspection focusing only on the parts with possible defects detected by AI reduced the inspection time by 80%
Inspection of Windmill Turbine Blade Siemens Gamesa Wind Power Inspection time: Reduction in 1.25 hours Efficiency improvement of 83%, Reduction for 34,000 hours a year FIAIR as AI package of Non-destruction inspection was released in Europe from September, 2017 Size and Number Length: 75m Speed: 100 km/h Issues Inspection time for 1 blade Non-destruction Test by Ultrasonic Scanning from 30 to 45 min. Inspection from Ultrasonic Image Highly-skilled inspectors check 1 blade from 6 to 8 hours Every year 5,000 Blades Inspection End Inspection from Ultrasonic Image Detection of Defects by Deep Learning Inspection time of Ultrasonic images Reduction in 1.25 hour 21 38 Technologies Manual inspection focusing only on the parts with possible defects detected by AI reduced the inspection time by 80%
Cyber Security: Malware Intrusion Detection Automated malware activities check by AI A few days by an expert A few minutes with high accuracy Deep Tensor learns multi attack patterns automatically Greatly improves malware detection accuracy Technology Effect Various Real data transaction compound SNS Communic ation Log transform Graph Data multi Tensor Deep Tensor Deep learning Learn multi attack patterns Unique nodes 343 Source IP address 249 Destination IP address Source Port No. 223 Destination Port No. Packet length Executed commands Attack exists. DT: correct, conventional: NG 23
Cyber Security: Malware Intrusion Detection Automated malware activities check by AI A few days by an expert A few minutes with high accuracy Deep Tensor learns multi attack patterns automatically Greatly improves malware detection accuracy Technology Effect Various Real data transaction compound SNS Communic ation Log transform Graph Data multi Tensor Deep Tensor Deep learning Learn multi attack patterns Unique nodes 343 Source IP address 249 Destination IP address Source Port No. 223 Destination Port No. Packet length Executed commands Attack exists. DT: correct, conventional: NG 23
Program for an Integrated Database of Clinical and Genomic Information National Project Funded by AMED in Japan Structuring a large number of academic articles and knowledge databases in the LOD format providing reliability evaluation and diagnostic support through inference Collecting various DBs and structuring knowledge Annotation support at 11 organizations Open to public A large number of academic articles and knowledge DBs scattered around the world PubMed ClinVar KEGG MEDICUS Literature DB Gene mutation DB Drug DB MedGen HGVD DrugBank Collect LOD utilisation platform Gene Protein Malignant Disease Drug Function LOD knowledge graph Inference of clinical interpretation Evidence output Candidate drug AI inference engine (based on machine learning) Cancer Dementia Pathway simulation Clinical/genomic information Rare/intractable diseases Infection Other Integrated Database of Clinical and Genomic Information Genomic medicine tailored for individual patients 24
Program for an Integrated Database of Clinical and Genomic Information National Project Funded by AMED in Japan Structuring a large number of academic articles and knowledge databases in the LOD format providing reliability evaluation and diagnostic support through inference Collecting various DBs and structuring knowledge Annotation support at 11 organizations Open to public A large number of academic articles and knowledge DBs scattered around the world PubMed ClinVar KEGG MEDICUS Literature DB Gene mutation DB Drug DB MedGen HGVD DrugBank Collect LOD utilisation platform Gene Protein Malignant Disease Drug Function LOD knowledge graph Inference of clinical interpretation Evidence output Candidate drug AI inference engine (based on machine learning) Cancer Dementia Pathway simulation Clinical/genomic information Rare/intractable diseases Infection Other Integrated Database of Clinical and Genomic Information Genomic medicine tailored for individual patients 24
Brewing of Japanese Sake Asahi Shuzo Co., Ltd., Collaboration of AI prediction with Japanese Sake of a special brand DASSAI Verification of practical use of AI in brewing of Japanese sake Joint Field Trial started in April, 2018 Issues Stable supply High-quality maintenance Visualization of know-how Temperature management Shortage in Experts by aging Management of volume of water Technologies Integration of 3 technologies Physics model Optimized control Machine learning Prediction by combining of physical model, calculation value, and measurement value Physics model's lack is supplemented Calculation for changes in temperature change and timing of addition water 25 Prediction results Achievement of ideal brewing equipment control by AI prediction
Brewing of Japanese Sake Asahi Shuzo Co., Ltd., Collaboration of AI prediction with Japanese Sake of a special brand DASSAI Verification of practical use of AI in brewing of Japanese sake Joint Field Trial started in April, 2018 Issues Stable supply High-quality maintenance Visualization of know-how Temperature management Shortage in Experts by aging Management of volume of water Technologies Integration of 3 technologies Physics model Optimized control Machine learning Prediction by combining of physical model, calculation value, and measurement value Physics model's lack is supplemented Calculation for changes in temperature change and timing of addition water 25 Prediction results Achievement of ideal brewing equipment control by AI prediction
Brewing of Japanese Sake Asahi Shuzo Co., Ltd., Collaboration of AI prediction with Japanese Sake of a special brand DASSAI Verification of practical use of AI in brewing of Japanese sake Joint Field Trial started in April, 2018 Issues Stable supply High-quality maintenance Visualization of know-how Temperature management Shortage in Experts by aging Management of volume of water Technologies Integration of 3 technologies Physics model Optimized control Machine learning Prediction by combining of physical model, calculation value, and measurement value Physics model's lack is supplemented Calculation for changes in temperature change and timing of addition water 25 Prediction results Achievement of ideal brewing equipment control by AI prediction
Scoring Support for Artistic Gymnastics Recognition of gymnastics performance and degree of perfection utilizing technology that catches person s movement accurately Scoring Support of upgrading gymnastics performance Issues Technologies Upgrading gymnastics performance Judgement becomes difficult every year 3D laser sensor technology Highly accurate laser control and Picture corner control technology Laser injection Luminescence circuit Skeleton recognition technology by AI The joint position is recognized in high-speed and high accuracy using 3D data Reflected ray Distance measurement circuit The angle of the beam is squeezed according to the distance when the person is detected. 26
Scoring Support for Artistic Gymnastics Recognition of gymnastics performance and degree of perfection utilizing technology that catches person s movement accurately Scoring Support of upgrading gymnastics performance Issues Technologies Upgrading gymnastics performance Judgement becomes difficult every year 3D laser sensor technology Highly accurate laser control and Picture corner control technology Laser injection Luminescence circuit Skeleton recognition technology by AI The joint position is recognized in high-speed and high accuracy using 3D data Reflected ray Distance measurement circuit The angle of the beam is squeezed according to the distance when the person is detected. 26
Closing
In the Digital Era 信頼と創造の富士通 Trust and Co-creation