Federated Learning: From Algorithms to System Implementation

Contents
Preface
About the Authors
Part I: Federated Learning Knowledge
	Chapter 1 Introduction to Federated Learning
		1.1 What Is Federated Learning
			1.1.1 History of Federated Learning
			1.1.2 Federated Learning Overview
			1.1.3 Classification of Federated Learning
		1.2 Current Application State and Development Trends of Federated Learning
			1.2.1 Current Applications of Federated Learning
			1.2.2 Development Trends of Federated Learning
		1.3 Distributed Machine Learning and Federated Learning
			1.3.1 History of Distributed Machine Learning
			1.3.2 Overview of Distributed Machine Learning
			1.3.3 The Co-development of Distributed Machine Learning and Federated Learning
		1.4 Conclusion
	Chapter 2 Federated Learning Application Scenarios
		2.1 Federated Learning and Finance
		2.2 Federated Learning and Biomedicine
		2.3 Federated Learning and Computer Vision
		2.4 Federated Learning and Natural Language Processing
		2.5 Federated Learning, Cloud Computing, and Edge Computing
		2.6 Federated Learning and Computer Hardware
		2.7 Conclusion
	Chapter 3 Common Privacy Protection Technologies
		3.1 Privacy Preserving Machine Learning
			3.1.1 Overview of Privacy Preserving Machine Learning
			3.1.2 Developments in Privacy-Preserving Machine Learning
		3.2 Commonly Used Privacy Protection Techniques
			3.2.1 Differential Privacy
				3.2.1.1 Introduction
				3.2.1.2 Differential privacy
				3.2.1.3 Applications of differential privacy
				3.2.1.4 Other algorithms
			3.2.2 Secure Multi-Party Computation
				3.2.2.1 Definition of secure multi-party computation
				3.2.2.2 Yao’s Garbled circuit
				3.2.2.3 Secret sharing
				3.2.2.4 Oblivious transfer
				3.2.2.5 Applications of secure multi-party computation
			3.2.3 Homomorphic Encryption
				3.2.3.1 Introduction
				3.2.3.2 Introduction to homomorphic encryption
				3.2.3.3 Partial homomorphic encryption (PHE)
				3.2.3.4 Somewhat homomorphic encryption (SWHE)
				3.2.3.5 Fully homomorphic encryption (FHE)
				3.2.3.6 Application of homomorphic encryption
				3.2.3.7 Implementation of homomorphic encryption
				3.2.3.8 Conclusion
Part II: Federated Learning Algorithms
	Chapter 4 Tree-Based Models in Vertical Federated Learning
		4.1 Introduction to Tree-Based Models
		4.2 Vertical Federated Random Forest Algorithm
			4.2.1 Algorithm Structure
				4.2.1.1 Plaintext and ciphertext transmission
			4.2.2 Algorithm Description
			4.2.3 Security Analysis
				4.2.3.1 Data transmission and statistics
				4.2.3.2 Security of labels
				4.2.3.3 Security of features
				4.2.3.4 Privacy-enhanced federated random forest algorithm
				4.2.3.5 Privacy-enhanced universal federated learning algorithm
		4.3 Vertical Federated Gradient Boosting Algorithm
			4.3.1 Review of XGBoost Algorithm
			4.3.2 Review of SecureBoost Algorithm
			4.3.3 Vertical Federated Gradient Boosting Algorithm
				4.3.3.1 Encrypted gradient sharing
				4.3.3.2 Model storage and inference
	Chapter 5 Vertical Federated Linear Regression Algorithm
		5.1 Vertical Federated Linear Regression
			5.1.1 Training Process
			5.1.2 Inference Process
			5.1.3 A Dilemma of Vertical Federated Learning
		5.2 Federal Multi-View Linear Regression
			5.2.1 Second-Order Optimization Method Based on BFGS
			5.2.2 Secure Computation Protocol
				5.2.2.1 Basic cryptographic components and assumptions
				5.2.2.2 Paillier algorithm based on distributed key generation and threshold decryption
				5.2.2.3 Training protocol
				5.2.2.4 Inference protocol
			5.2.3 Conclusion
	Chapter 6 Vertically Federated Kernel Learning
		6.1 Introduction
		6.2 Brief Review of Doubly Stochastic Kernel Methods
			6.2.1 Problem Statement
			6.2.2 A Brief Introduction to Kernel Methods
			6.2.3 Random Fourier Features Approximation
			6.2.4 Doubly Stochastic Gradient
		6.3 Vertical Federated Kernel Learning Approach
			6.3.1 Problem Statement
			6.3.2 Algorithm Structure
			6.3.3 Algorithm Design
			6.3.4 Scenario Case Study
		6.4 Theoretical Analysis
			6.4.1 Convergence Analysis
			6.4.2 Security Analysis
			6.4.3 Complexity Analysis
		6.5 Experiments
			6.5.1 Experimental Setup
			6.5.2 Results and Discussions
		6.6 Conclusion
	Chapter 7 Asynchronous Vertical Federated Learning Algorithm
		7.1 Introduction
		7.2 Related Work
			7.2.1 Overview of Existing Work
			7.2.2 Review of SGD-Type Algorithms
		7.3 Problem Statement
		7.4 Asynchronous Vertical Federated Learning Algorithm
			7.4.1 Algorithm Framework
			7.4.2 Algorithm Description
			7.4.3 Scenario Case Study
		7.5 Theoretical Analysis
			7.5.1 Convergence Analyses
			7.5.2 Security Analysis
			7.5.3 Complexity Analysis
		7.6 Experimental Results
			7.6.1 Experimental Setup
			7.6.2 Results and Discussions
		7.7 Conclusion
	Chapter 8 Secure Bilevel Asynchronous Vertical Federated Learning with Backward Updating
		8.1 Introduction
		8.2 Problem Formulation
		8.3 Understanding Algorithms
			8.3.1 VFB2 Framework
			8.3.2 Details of Algorithms
			8.3.3 Scenario Case Study
		8.4 Theoretical Analysis
			8.4.1 Convergence Analysis for Strongly Convex Problem
			8.4.2 Convergence Analysis for Non-convex Problem
			8.4.3 Security Analysis
		8.5 Experiments
			8.5.1 Experiment Settings
			8.5.2 Evaluations of Asynchronous Efficiency and Scalability
		8.6 Conclusion
	Chapter 9 Vertical Federated Deep Learning Algorithms
		9.1 Introduction
		9.2 Vertical Federated Deep Learning Algorithms
			9.2.1 Algorithm Framework
			9.2.2 Algorithm Details
			9.2.3 Scenario Case Study
		9.3 Theoretical Analysis
			9.3.1 Complexity Analysis
			9.3.2 Security Analysis
		9.4 Experiments
			9.4.1 Experimental Setup
			9.4.2 Results and Discussion
		9.5 Conclusion
	Chapter 10 Faster Secure Data Mining Framework via Homomorphic Encryption
		10.1 Introduction
		10.2 Related Work
		10.3 Faster Secure Data Mining Framework via Homomorphic Encryption
			10.3.1 Algorithm Framework
			10.3.2 Algorithm Description
		10.4 Experimental Results
			10.4.1 Distributed Learning
			10.4.2 Federated Learning
		10.5 Conclusion
	Chapter 11 Horizontal Federated Learning Algorithms
		11.1 Introduction to Horizontal Federated Learning
		11.2 Common Distributed Optimization Algorithms
			11.2.1 Synchronous Parallel Algorithms
			11.2.2 Asynchronous Parallel Algorithms
		11.3 Synchronous Horizontal Federated Learning Algorithms
		11.4 Asynchronous Horizontal Federated Learning Algorithms
		11.5 Rapid Communication in Horizontal Federated Learning Algorithms
		11.6 Conclusion
	Chapter 12 Mixed Federated Learning Algorithms
		12.1 Scenario Requirements for Mixed Federated Algorithms
		12.2 Algorithm Details
			12.2.1 Algorithm Design
				12.2.1.1 Gradient update
				12.2.1.2 Mixed node splitting
			12.2.2 Model Storage and Mixed Inference
				12.2.2.1 Storing mixed federal gradient boosted trees
				12.2.2.2 Optimized inference process
		12.3 Background of the Mixed Federated Linear Regression Algorithm
		12.4 Objective Function of Mixed Linear Regression
		12.5 Improvements in First and Second Derivative Calculations
			12.5.1 For Private Data
			12.5.2 For ID-Overlapping Data
				12.5.2.1 Gradient
				12.5.2.2 Hessian matrix
		12.6 Secure Computation Protocol
			12.6.1 Key Generation Based on Secret Sharing
			12.6.2 Homomorphic Encryption with Threshold Decryption
			12.6.3 Secure Federated Training Method
	Chapter 13 Federated Reinforcement Learning
		13.1 Overview of Reinforcement Learning
			13.1.1 Markov Property
			13.1.2 Various Types of Policies
			13.1.3 Expected Returns
			13.1.4 Different Components and Setups for Learning Policies
		13.2 Introduction to Reinforcement Learning Algorithms
			13.2.1 Value-Based RL
			13.2.2 Fitted Q-Learning Algorithm
			13.2.3 Deep Q-Networks
			13.2.4 Double DQN
			13.2.5 Policy-Based RL
			13.2.6 Stochastic Policy Gradient
			13.2.7 Deterministic Policy Gradient
			13.2.8 Actor–Critic Method
			13.2.9 Model-Based RL
		13.3 Distributed and Federated Reinforcement Learning
			13.3.1 Distributed Reinforcement Learning
			13.3.2 Federated Reinforcement Learning
		13.4 Conclusion
Part III: Federated Learning Systems
	Chapter 14 Detailed Exploration of the FedLearn Federated Learning System
		14.1 Open-Source Federated Learning Systems and Their Pain Points
			14.1.1 Programming Language and Environment
			14.1.2 Big Data and Computational Efficiency
		14.2 Advantages of the FedLearn Federated Learning System
		14.3 FedLearn System Architecture Design
			14.3.1 Common Federated Learning System Architectures
			14.3.2 Overview of FedLearn Architecture
			14.3.3 FedLearn Standard Architecture Features
			14.3.4 Distributed Federated Learning
			14.3.5 Blockchain Federated Learning Architecture
		14.4 Cross-Language Support in FedLearn
			14.4.1 Introduction to Remote Procedure Call Technology
				14.4.1.1 Calling a local function
				14.4.1.2 Remote procedure calls
				14.4.1.3 New challenges with remote procedure calls
				14.4.1.4 Do RPCs have advantages?
		14.5 High-Performance Open-Source RPC Framework: gRPC
			14.5.1 Unique Advantages of gRPC
				14.5.1.1 Cross-language invocation
				14.5.1.2 Open source and standardization
				14.5.1.3 Communication security
			14.5.2 Key Concepts of gRPC
				14.5.2.1 Protocol buffers
				14.5.2.2 gRPC services
				14.5.2.3 gRPC compilation and API content population
				14.5.2.4 Standard RPC lifecycle
				14.5.2.5 Special RPC lifecycle
				14.5.2.6 Protocol buffers versioning
		14.6 Decoupling of FedLearn System Services and Algorithms
			14.6.1 Phase Concept and Automata
			14.6.2 Componentization
			14.6.3 Other System-Level Optimizations
		14.7 FedLearn Deployment and Use
			14.7.1 System Components and Functionality
				14.7.1.1 Coordinator
				14.7.1.2 Client
				14.7.1.3 Frontend
			14.7.2 Standard Deployment
				14.7.2.1 Coordinator deployment
				14.7.2.2 Client deployment
				14.7.2.3 Front-end deployment
			14.7.3 Deployment for Distributed Version
			14.7.4 Container-Based Deployment
			14.7.5 User Interface Operations and API
				14.7.5.1 System permissions module
				14.7.5.2 Project module
				14.7.5.3 Training management module
				14.7.5.4 Inference module
				14.7.5.5 Model application module
	Chapter 15 Application of gRPC in FedLearn
		15.1 Application Example I: Vertical Federated Random Forest Algorithm
			15.1.1 Overview of the Algorithm Process
				15.1.1.1 Designing gRPC function interfaces
				15.1.1.2 Interface scalability
				15.1.1.3 Plaintext and ciphertext transfers
				15.1.1.4 Data scale and dimensions
				15.1.1.5 Automated framework generation and logic populating
		15.2 Application Example II: Horizontal Federated Learning Scenario
			15.2.1 Brief on Horizontal Federated Learning
			15.2.2 Framework Design and Implementation of Horizontal Federated Learning in FedLearn
				15.2.2.1 FedLearn horizontal federated learning framework
				15.2.2.2 FedLearn horizontal federated learning control flow
				15.2.2.3 gRPC transmitted message payloads
			15.2.3 Utilizing gRPC to Support Diverse Model Types
				15.2.3.1 Unified command dispatching on the Python side
				15.2.3.2 Inheritance relationships of models on the Python side
				15.2.3.3 Supporting model parameter transmission across different network frameworks through gRPC byte streams
	Chapter 16 Performance Optimization Practices in Real-World Scenarios
		16.1 Introduction to FedLearn Business Scenarios
			16.1.1 Precise Marketing Monitoring of Financial Products
			16.1.2 Intelligent Credit Scoring
		16.2 From 0-1: Practical Optimization of Federated Learning Algorithms
		16.3 Performance Optimization
			16.3.1 GMP Computing Library
			16.3.2 Accelerating Homomorphic Encryption Computations with the GMP Library in Python
				16.3.2.1 Accelerating homomorphic encryption computations with the GMP library in Java
			16.3.3 Homomorphic Encryption Computation Protocol Optimization
		16.4 Optimization of Engineering Service Performance
			16.4.1 Parallel Optimization
				16.4.1.1 Parallel optimization in Java
				16.4.1.2 Parallel optimization in Python
		16.5 Optimizing Information Transfer across Multiple Machines
			16.5.1 Streamlining Data Transfer
			16.5.2 Communication Message Compression and Asynchronous Communication
		16.6 Real-Time Inference Optimization
			16.6.1 Conclusion
	Chapter 17 Federated Learning Based on Blockchain
		17.1 Introduction to Blockchain
			17.1.1 Overview of Blockchain
				17.1.1.1 What is blockchain
				17.1.1.2 Development of blockchain
				17.1.1.3 Types of blockchains
			17.1.2 Technical Features of Blockchain
				17.1.2.1 Cryptography
				17.1.2.2 Consensus algorithm
				17.1.2.3 Decentralized network
				17.1.2.4 Smart contracts
			17.1.3 Integration of Blockchain Technology with Modern Technologies
		17.2 Innovative Integration of Federated Learning and Blockchain
			17.2.1 Innovations in Architecture
			17.2.2 Innovations in Process
			17.2.3 Data Support
			17.2.4 Incentive Support
			17.2.5 Regulatory and Audit Support
		17.3 Blockchain-Based Federated Learning Incentive Algorithms
			17.3.1 Model Quality Assessment Incentive Algorithm
				17.3.1.1 Initialization phase
				17.3.1.2 Local training phase
				17.3.1.3 Evaluation phase
			17.3.2 Weighted Incentive Algorithm
				17.3.2.1 Blockchain-based weighted incentive design
				17.3.2.2 Weight value calculation
			17.3.3 Incentive Allocation Algorithm
		17.4 Implementation of a Blockchain-Based Federated Learning System
			17.4.1 System Model
				17.4.1.1 Trustworthy collaborative network
				17.4.1.2 Poisoning attacks
				17.4.1.3 Inference attacks
				17.4.1.4 Model and workload evaluation
				17.4.1.5 Identity management
				17.4.1.6 Introducing audit and regulation
			17.4.2 System Architecture
				17.4.2.1 Component architecture
				17.4.2.2 Process architecture
				17.4.2.3 Deployment architecture
Bibliography
Index