XMANAI Technical Project Overview
Download as PPTX, PDF•0 likes•371 views
This document provides an overview of an AI project called XMANAI. It discusses: 1) Key AI concepts like definitions, policies, trends and explainable AI. 2) The challenges XMANAI aims to address like increasing trust and transparency in AI for manufacturing. 3) XMANAI's vision to develop explainable hybrid and graph AI models coupled with complex data and model management to solve manufacturing problems.
XMANAI Technical Project Overview
- 1. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 957362 Technical Project Overview 2020-11-10 Kick-off Meeting Dr. Fenareti Lampathaki (Suite5) – Technical Coordinator
- 2. TABLE OF CONTENTS 01 02 03 04 AI Basics Definition, Policies, Trends, Explainable AI, Technologies Rationale & Motivation Key Challenges Vision Approach, Concept XMANAI in Perspective Methodology, Roadmap 2020-11-01 Kick-off Meeting 2
- 3. AI Basics 01 Definition, Policies, Trends, Explainable AI, Technologies 3 2020-11-01 Kick-off Meeting
- 4. • Artificial intelligence (AI) refers to systems that display intelligent behaviour by analysing their environment and taking actions – with some degree of autonomy – to achieve specific goals. • AI-based systems can be: • Software-based, acting in the virtual world (e.g. voice assistants, image analysis software, search engines, speech and face recognition systems) • Embedded in hardware devices (e.g. advanced robots, autonomous cars, drones or Internet of Things applications) Artificial Intelligence 2020-11-01 Kick-off Meeting 4 Sources: EC (2018) Communication from the Commission: Artificial Intelligence for Europe. SWD(2018) 137 final & https://qbi.uq.edu.au/brain/intelligent-machines/history-artificial-intelligence
- 5. 5 2020-11-01 Kick-off Meeting AI at EU Policy Level A Common European industrial (manufacturing) data space, to support the competitiveness and performance of the EU’s industry, allowing to capture the potential value of use of non-personal data in manufacturing (estimated at € 1,5 trillion by 2027).
- 6. 6 2020-11-01 Kick-off Meeting AI Taxonomies Thematic subdomains Natural Language Processing (NLP) & Generation(NLG) Computer vision AI Applications, Infrastructure, Platforms, Software as a Service (AIaaS, IaaS, PaaS, SaaS) Machine learning (ML) methods Robotics & Automation Processes Connected & Automated Vehicles (CAVs) AI thematic subdomains and top-10 terms by relevance to the topic, of industrial and R&D activities, 2009-2018 – Source: https://doi.org/10.1016/j.telpol.2020.101943 + Based on breadth of intelligence, learning ability, type of application, learning paradigm, etc.
- 7. 7 2020-11-01 Kick-off Meeting Trends in AI, Data Science & ML 2020 • Highlights from Hype Cycle for Data Science and Machine Learning, 2020: • On the Rise: Self-Supervised Learning - Federated Machine Learning – Kubeflow - Transfer Learning • At the Peak: Data Labeling and Annotation Services - Explainable AI – MLOps - Augmented DSML - AutoML - Deep Neural Networks (Deep Learning) - Prescriptive Analytics • Sliding Into the Trough: Graph Analytics - Advanced Video/Image Analytics - Event Stream Processing • Climbing the Slope: Predictive Analytics - Text Analytics • Entering the Plateau: Apache Spark - Notebooks
- 8. • Explainability that reflects the ability to understand and explain in human terms what is happening with an AI model and how it works under the hood, promoting inspection and traceability of actions undertaken. • Interpretability that refers to the degree to which a human can observe cause-and-effect situations, understand the cause of a decision and anticipate how changes in the data or the model will alter the results. • Trustability that embraces the ability of AI models to provide accurate, trustworthy and performant predictions. Explainable AI (XAI) 2020-11-01 Kick-off Meeting 8 GDPR AI: It’s a machine failure or not? XAI: Why it’s a machine failure? Source: https://www.cc.gatech.edu/~alanwags/DLAI2016/(Gunning)%20IJCAI-16%20DLAI%20WS.pdf
- 9. Explainability Aspects 2020-11-01 Kick-off Meeting 9
- 10. 10 2020-11-01 Kick-off Meeting AI Technologies Landscape 2020
- 11. 11 2020-11-01 Kick-off Meeting AI in Manufacturing AI potential across the breadth and depth of manufacturing operations Sources: https://www.capgemini.com/wp-content/uploads/2019/12/AI-in-manufacturing- operations.pdf & https://www.pwc.com/gx/en/industrial-manufacturing/pdf/intro-implementing-ai- manufacturing.pdf
- 12. Rationale & Motivation 02 Key Challenges 12 2020-11-01 Kick-off Meeting
- 13. 13 2020-11-01 Kick-off Meeting Key Business Challenges “Why did the AI system make a specific prediction or decision?” “Why didn’t the AI system decide something else?“ “When did the AI system succeed and when did it fail and what was the impact?” How to avoid undetected bias, mistakes, and miscomprehensions creeping into decision-making? How to facilitate robustness, accuracy and performance? How to ensure fair decision making? How to provide really actionable insights? I. How to Increase Human Trust in AI II. How to Increase Transparency and Reliability of AI “The AI made us take this decision, not sure why…”
- 14. 14 2020-11-01 Kick-off Meeting Just some of the key “Data/AI” Challenges I. Efficient and Secure Data Management II. Traceable AI Model Lifecycle Management III. Trusted Data and AI Model Sharing Inconsistent, incomplete or missing data with low dimensionality -> overfitting or underfitting AI models Properly preparing and manipulating the data (even 80% of time in AI projects!) Packaging, deploying and scaling AI models/pipelines in different execution environments in an interoperable manner Keeping track of AI experiments and reproducing code and results Transfer learning Collaboration between data scientists, engineers and business experts IPR-compliant “assets” exchange Equilibrium point for AI-related assets handling AI’s “transparency paradox” & Ethics
- 16. 16 2020-11-01 Kick-off Meeting Vision at a glance… • WHY? For optimizing performance and manufacturing products’ and processes’ quality… For accurately forecasting product demand… For production optimization and predictive maintenance… For enabling agile planning processes… For understanding and cultivating different skills that are required in manufacturing in order to transition to the AI era “XMANAI aims at placing the indisputable power of Explainable AI at the service of manufacturing and human progress, carving out a “human-centric”, trustful approach that is respectful of European values and principles, and adopting the mentality that “our AI is only as good as we are”. XMANAI will help the manufacturing value chain to shift towards the amplifying AI era by coupling (hybrid and graph) AI "glass box" models that are explainable to a "human-in-the-loop" and produce value-based explanations, with complex AI assets (data and models) management-sharing-security technologies to multiply the latent data value in a trusted manner, and targeted manufacturing apps to solve concrete manufacturing problems with high impact. ” Target Users: Data Scientists & Data Engineers; Business Users in Manufacturing
- 17. Explainable AI Circles in XMANAI 2020-11-01 Kick-off Meeting 17 AI Axis I. Basic Analytics AI Axis II. Machine Learning AI Axis III. Deep Learning Graph AI Algorithms - Understand AI Models - Hybrid AI Algorithms - Understand AI Results - Traditional AI Algorithms – Understand Data - Multi-Party Asset Contracts for acquiring AI Models, Explanations, Results and / or (Missing) Industrial Data Trust Level 1 - Emerging XAI Circle Trust Level 2 - Developing XAI Circle Trust Level 3 - Established XAI Circle Sample Data Exploration Features Visualizations Elicited Explanations Features Surrogate Models Directly Interpretable Models Explanation Interfaces Knowledge Graphs -> Graph Feature Engineering -> Graph Native Learning
- 18. 18 2020-11-01 Kick-off Meeting XAI Models Portfolio in XMANAI OBJ.1: >30 hybrid AI baseline models. >15 graph AI baseline models. >16 hybrid trained AI models. >8 trained graph AI models. >4 surrogate models to increase explainability for hybrid AI models. >4 manufacturing problems addressed ΧΜΑΝΑΙ Catalogue of Explainable AI models Hybrid ML/DL Algorithms Graph ML/DL Algorithms Baseline Models Hybrid ML/DL Algorithms Graph ML/DL Algorithms Trained Models Surrogate Models: SHapley Additive exPlanations (SHAP) Local Interpretable Model-agnostic Explanations (LIME) Causal Models to Explain Learning (CAMEL) …
- 19. XMANAI Journey though the “Business Expert” Perspective 2020-11-01 Kick-off Meeting 19 Business Expert Perspective Extract Data Data At rest Data In motion Data Ingestion Data Mapping and Semantic Annotation Data Cleaning Data Provenance Data Update Management XMANAI Data/Model Management Bundle Safeguard Data Data Security and Privacy Assurance Data Access Policies Management Data Licensing Secure Data Storage in multiple modalities XMANAI Data/Model Management Bundle XMANAI Secure Asset Sharing Bundle Share / Acquire (Missing) Data Bilateral Asset Contract… Creation Approval Negotiation Signature Lifecycle Management XMANAI Core AI Bundle Understand Data Exploratory Data Analysis Data Modelling Data Curation Knowledge Graph Generation Manufacturing Systems & XMANAI Demonstrators Applications Interactive Results’ Visualizations Results and Explanations Extraction Notifications XMANAI AI Insights Bundle Derive Intelligence AI Results
- 20. 20 2020-11-01 Kick-off Meeting XMANAI Journey though the “Data Scientist” Perspective Data Scientist Perspective XMANAI Core AI Bundle Understand Data Exploratory Data Analysis Data Modelling Data Curation Knowledge Graph Generation XMANAI Core AI Bundle Manipulate Data Data Transformation Data Linking / Merging Graph Embeddings Dimensionality Reduction Feature Storage XMANAI Secure Asset Sharing Bundle Acquire (Missing) Data Bilateral Asset Contract… Creation Approval Negotiation Signature Lifecycle Management Evaluate the AI Model Cross Validation Model and Experiment Explanation Expert Collaboration and Validation of (Visualized) Results Model Safeguarding and / or Sharing XMANAI AI Insights Bundle XMANAI Core AI & AI Insights Bundles Fit the AI Model Training Feature Tuning Experiment Planning Experiment Execution Surrogate Models Execution Experiment Tracking XMANAI Core AI & AI Insights Bundles AI/ML Pipeline Definition Feature Engineering Surrogate Models Definition Models Storage Build an AI Model
- 21. 21 2020-11-01 Kick-off Meeting XMANAI Journey though the “Data Engineer” Perspective XMANAI Core AI & AI Insights Bundles Fit the AI Model Training Feature Tuning Experiment Planning Experiment Execution Surrogate Models Execution Experiment Tracking XMANAI Core AI & AI Insights Bundles AI/ML Pipeline Definition Feature Engineering Surrogate Models Definition Models Storage Build an AI Model Data Engineer Perspective Model Packaging Model Deployment Model Immediate / Scheduled Execution Model Scaling XMANAI Core AI Bundle Deploy the AI Model Manufacturing Systems & XMANAI Demonstrators Applications AI Models
- 22. 22 2020-11-01 Kick-off Meeting XMANAI Architecture Core AI Management Platform Open APIs Secure Execution Clusters (SEC) Process Optimization App XMANAI ON-PREMISE ENVIRONMENTS Stakeholders’ On-Premise Environment (OPE) XMANAI CLOUD INFRASTRUCTURE XMANAI MANUFACTURING APPS PORTFOLIO Product Demand Forecasting App Process/Product Quality Optimization App Process Optimization & Semi- Autonomous Planning App Services in a nutshell: • Data & Models Collection Services • Scalable Storage Services • Data Manipulation Services • AI Model Lifecycle Services • AI Execution Services • AI Insights Services • Secure Asset Sharing Services • Data & Models Governance Services
- 23. AI for Production Optimization AI for Product Demand Planning AI for Process/Product Quality Optimization AI for Smart Semi- autonomous Hybrid Measurement Planning 23 2020-11-01 Kick-off Meeting XMANAI Demonstrators
- 24. XMANAI in Perspective 04 Methodology, Roadmap 24 2020-11-01 Kick-off Meeting
- 25. XMANAI Methodology 2020-11-01 Kick-off Meeting 25 Analyzing the Manufacturing Industry Needs Early Preparing Market Entry Developing the XMANAI Platform State-of-play analysis, Ethics and human aspects in decision making and AI Industrial needs & barriers identification Business scenarios and requirements consolidation Research agenda elaboration, Minimum Viable Product (MVP) conceptualization Market analysis Detailed exploitation strategy New AI-enabled business models Business model and plan Business cases from pilot experience, Scale-up, Transfer Learning and Replication Dissemination, Communication & Marketing Stakeholder engagement Industrial data management, sharing and AI models lifecycle management methods elaboration Manufacturing data in depth- exploration Baseline and trained XAI models for manufacturing Architectural design Development of XMANAI Data Services Bundles Integration of the XMANAI core platform & on-premise environm. I IV II Verifying, Validating and Demonstrating the XMANAI Platform XAI Models Evaluation XAI Models Ethics and Security assessment Project Verification & Validation framework elaboration Platform Technical Verification and Validation Demonstrators Preparation and Implementation Business Validation Impact Assessment & Lessons learnt III
- 26. Project “Technical” Roadmap 2020-11-01 Kick-off Meeting 26 M2 M37 M6 M8 M9 M11 Updated MVP Asset Mgmt Bundles-1st Rel & AI Bundles-1st Rel Eval Plan M3 M40 M39 M38 M42 M4 M5 M10 M7 M12 M23 M19 M17 M16 M14 M22 M21 M20 M15 M18 M13 M24 M26 M30 M32 M33 M35 M27 M28 M29 M34 M31 M36 M25 M41 Platform-Alpha Version Final AI Models Platform v1.0 Demo Results-Beta Phase & Impact Assess. Concept & MVP Architecture Asset Mgmt Bundles Design AI Bundles Design Final MVP SoTA Demonstrators Reqs Asset Mgmt Bundles-2nd Rel AI Bundles-2nd Rel Demo Results-Alpha Phase Draft AI Catalogue AI Models-1st Round AI Models-2nd Round Platform-Beta Version Updated Demo Reqs
- 27. • “Research & Innovation”… Expectations are really high ! • High-quality results must be submitted to the EC on time • Teamwork… Clear roles and commitments already in the XMANAI consortium • Grasp the opportunities to innovate, align to the actual needs and make a tangible impact in industry 27 2020-11-01 Kick-off Meeting Key Messages I. Without Data, there is no AI ! II. Without AI explainability, there is no trust ! III. Without AI ethics, there is no adoption ! If humans do not understand why/how a decision/prediction is reached, they shall not adopt/enforce it… MAKING AI UNDERSTANDABLE !
- 28. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 957362 Thank your for your attention! 28 [email protected]
- 29. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 957362 www.ai4manufacturing.eu [email protected] /XMANAI /XMANAI /XMANAI 29 Dr. Fenareti Lampathaki Technical Director Suite5 Data Intelligence Solutions Limited www.suite5.eu [email protected] /fenareti /fenareti.lampathaki / fenareti.lampathaki