This book provides an overview of Federated Learning and how it can be used to build real-world AI-enabled applications. Real-world AI applications frequently have training data distributed in many different locations, with data at different sites having different properties and different formats. In many cases, data movement is not permitted due to security concerns, bandwidth, cost or regulatory restriction. Under these conditions, techniques of federated learning can enable creation of practical applications. Creating practical applications requires implementation of the cycle of learning from data, inferring from data, and acting based on the inference. This book will be the first one to cover all stages of the Learn-Infer-Act cycle, and presents a set of patterns to apply federation to all stages. Another distinct feature of the book is the use of real-world applications with an approach that discusses all aspects that need to be considered in an operational system, including handling of data issues during federation, maintaining compliance with enterprise security policies, and simplifying the logistics of federated AI in enterprise contexts. The book considers federation from a manner agnostic to the actual AI models, allowing the concepts to be applied to all varieties of AI models. This book is probably the first one to cover the space of enterprise AI-based applications in a holistic manner.
Author(s): Dinesh C. Verma
Publisher: CRC Press
Year: 2021
Language: English
Pages: 218
City: Boca Raton
Cover
Title Page
Copyright Page
Dedication
Preface
Table of Contents
1. Introduction to Artificial Intelligence
1.1 Business Operations Model
1.2 The Learn→Infer→Act Cycle
1.3 Process of Creating an AI Model
1.4 The Role of Humans in AI
1.4.1 Human Role in Learn Phase
1.4.2 Human Role in Infer and Act Phase
1.5 AI Model Representation
1.5.1 Functional Representation
1.5.2 Decision Tables
1.5.3 Decision Trees
1.5.4 Rule Sets
1.5.5 Neural Networks
1.5.6 Matrix Transform-based Representation
1.5.7 Distance-based Models
1.5.8 Finite State Machine Models
1.5.9 Equivalence of AI Models
1.6 Model Learning Approaches
1.7 Spatial and Temporal Aspects of Learning
1.8 AI Enabled Functions
1.8.1 Classification
1.8.2 Clustering
1.8.3 Anomaly Detection
1.8.4 Mapping
1.8.5 Filtering
1.8.6 Function Modeling
1.8.7 Goal Attainment
1.9 Summary
2. Scenarios for Federated AI
2.1 Abstracted Patterns for Enterprise AI
2.1.1 Centralized AI
2.1.2 Edge Inference
2.1.3 Federated Edge Inference
2.1.4 Edge Learning
2.1.5 Proxy Learning
2.2 Motivations for Federated Learning
2.2.1 Operational Cost
2.2.2 Network Constraints
2.2.3 Data Privacy Regulations
2.2.4 Security and Trust Considerations
2.3 Consumer and Enterprise Federated Learning
2.3.1 Consumer Federated Learning
2.3.2 Enterprise Federated Learning
2.4 Enterprise Federated Learning Scenarios
2.4.1 Subsidiaries and Franchises
2.4.2 Mergers and Acquisitions
2.4.3 Outsourced Operations
2.4.4 Telecommunications Networks
2.4.5 Consortia and Coalitions
2.4.6 Regulated Industries
2.5 Summary
3. Naive Federated Learning Approaches
3.1 Federated Learning of Metrics
3.2 Function Estimation
3.3 Federated Learning for Function Estimation
3.4 Federated Learning for Neural Networks
3.5 Federation of Miscellaneous Models
3.5.1 Distributed Incremental Learning
3.6 Assumptions in Naive Federated Learning
3.7 Summary
4. Addressing Data Mismatch Issues in Federated AI
4.1 Converting to Common Input Format
4.1.1 Raw Data Types
4.1.2 Featured Data
4.2 Resolving Value Conflicts
4.2.1 Committee Approach to Reconciliation
4.2.2 Summarization Approach to Reconciliation
4.2.3 Cross-Site Confusion Matrix
4.2.4 Feature Space Analysis
4.3 Eliminating Poor Quality and Low Value Data
4.3.1 Reputation-based Data Selection
4.3.2 Value-based Data Selection
4.3.3 Policy-based Quality Improvement
4.4 Summary
5. Addressing Data Skew Issues in Federated Learning
5.1 Impact of Partitioned and Unbalanced Data
5.1.1 Data Skew Issues in Function Estimation
5.1.2 Label Partitioning Issues in Classification
5.2 Limited Data Exchange
5.3 Policy-based Ensembles
5.4 Summary
6. Addressing Trust Issues in Federated Learning
6.1 Scenarios with Multiple Trust Zones
6.1.1 Cloud-based Fusion Server
6.1.2 Multi-tenancy Cloud Site
6.1.3 Consortia and Alliances
6.1.4 Military Coalitions
6.2 Trust Zone Configurations
6.2.1 Trusted Fusion Server with Untrusted Fusion Clients
6.2.2 Untrusted Fusion Server with Trusted Fusion Clients
6.2.3 Untrusted Fusion Server with Untrusted Fusion Clients
6.3 Addressing Trust Issues with Business Arrangements
6.4 Addressing Trust Issues with Infrastructure Technology
6.5 Auditing and Logging
6.6 Encryption-based Approaches
6.6.1 Fully Homomorphic Encryption
6.6.2 Partial Homomorphic Model Learning
6.7 Differential Privacy-based Approaches
6.8 Summary
7. Addressing Synchronization Issues in Federated Learning
7.1 Overview of Synchronization Issues
7.2 Asynchronous Data Mismatch Issues
7.3 Ensemble-based Approaches
7.4 Conversion to Rule-based Models
7.5 Data Generator-based Model Fusion
7.6 Summary
8. Addressing Vertical Partitioning Issues in Federated Learning
8.1 General Approaches for Handling Vertical Partitioning
8.2 Rule-based Approach
8.3 Feature Prediction Approach
8.4 Feature Mapper Augmentation
8.5 Federated Inference
8.6 Summary
9. Use Cases
9.1 Collaborative Scam Detection
9.1.1 Collaboration within a Single Industry
9.1.2 Collaboration Across Industries
9.1.3 Effectiveness
9.2 Federated Network Management
9.3 Retail Coupon Recommendation
9.4 Summary
Appendix 1: Frameworks for Federated Learning
Appendix 2: Adversarial Federated Learning
References
Index
