Communication Efficient Federated Learning for Wireless Networks

Preface
Acknowledgements
Contents
1 Introduction
	1.1 Motivation of Distributed Learning
	1.2 Challenges of Deploying Distributed Learning
	1.3 Potential Techniques for Deploying Federated Learning
	1.4 Summary and Book Overview
2 Fundamentals and Preliminaries of Federated Learning
	2.1 Preliminaries of FL
		2.1.1 Common Federated Learning
		2.1.2 Federated Multi-Task Learning
		2.1.3 Model Agnostic Meta Learning Based FL
	2.2 Performance Metrics of FL Over Wireless Networks
		2.2.1 Training Loss
		2.2.2 Convergence Time
		2.2.3 Energy Consumption
		2.2.4 Reliability
	2.3 Effects of Wireless Factors on FL Metrics
	2.4 Research Directions of Deploying FL Over Wireless Networks
		2.4.1 Wireless Resource Management
		2.4.2 Compression and Sparsification
		2.4.3 FL with Over-the-Air Computation
		2.4.4 FL Training Method Design
		2.4.5 Industry Interest
3 Resource Management for Federated Learning
	3.1 Resource Management for FL Training Loss Minimization
		3.1.1 Wireless FL Model
			3.1.1.1 FL Parameter Transmission Model
			3.1.1.2 FL Parameter Error Rates
			3.1.1.3 Energy Consumption Model
		3.1.2 Problem Formulation
		3.1.3 Analysis of the FL Convergence Rate
		3.1.4 Optimization of FL Training Loss
			3.1.4.1 Optimal Transmit Power
			3.1.4.2 Optimal Uplink Resource Block Allocation
		3.1.5 Simulation Results and Analysis
	3.2 Resource Management for FL Convergence Time Minimization
		3.2.1 Wireless FL Model
		3.2.2 Problem Formulation
		3.2.3 Minimization of FL Convergence Time
			3.2.3.1 Gradient Based User Association Scheme
			3.2.3.2 Optimal RB Allocation Scheme
			3.2.3.3 Prediction of the Local FL Models
		3.2.4 Simulation Results and Analysis
	3.3 Resource Management for Energy Efficiency Optimization
		3.3.1 Wireless FL Model
			3.3.1.1 Local Computation
			3.3.1.2 Wireless Transmission
			3.3.1.3 GLobal FL Model Generation and Broadcast
		3.3.2 Problem Formulation
		3.3.3 Iterative Algorithm
	3.4 Conclusions
4 Quantization for Federated Learning
	4.1 Univrersal Vector Quantization for Federated Learning
		4.1.1 Wireless FL Model
		4.1.2 Problem Formulation
		4.1.3 Universal Vector Quantization based FL
		4.1.4 Performance Analysis
			4.1.4.1 Local SGD
			4.1.4.2 Quantization Error Bound
			4.1.4.3 FL Convergence Analysis
		4.1.5 Numerical Evaluations
			4.1.5.1 Quantization Error
			4.1.5.2 FL Convergence
	4.2 Variable Bitwidth Federated Learning
		4.2.1 Variable Bitwidth FL Model
			4.2.1.1 Training Process of Bitwidth Federated Learning
			4.2.1.2 Training Delay of Bitwidth Federated Learning
		4.2.2 Problem Formulation
		4.2.3 Optimization Methodology
			4.2.3.1 Components of Model Based RL Method
			4.2.3.2 Calculation of State Transition Probability
			4.2.3.3 Optimization of Device Selection and Quantization Scheme
		4.2.4 Numerical Evaluation
			4.2.4.1 Datasets and ML Models
			4.2.4.2 Convergence Performance Analysis
	4.3 Conclusions
5 Federated Learning with Over the Air Computation
	5.1 AirComp Principle and Techniques
		5.1.1 AirComp Principle
		5.1.2 Broadband AirComp
		5.1.3 MIMO AirComp
		5.1.4 Design of AirComp Federated Learning
			5.1.4.1 Model Update Distortion
			5.1.4.2 Device Scheduling
			5.1.4.3 Coding Against Interference
			5.1.4.4 Power Control
	5.2 Power Control Optimization for AirComp FL
		5.2.1 AirComp FL Model
		5.2.2 AirComp FL Convergence Analysis
			5.2.2.1 Basic Assumptions on Learning Model
			5.2.2.2 Optimality Gap Versus Aggregation Errors
			5.2.2.3 Optimality Gap Versus Transmission Power Control
			5.2.2.4 Convergence Analysis for AirComp-FL in Case I
			5.2.2.5 Convergence Analysis for AirComp-FL in Case II
		5.2.3 Power Control Optimization
			5.2.3.1 Power Control Optimization for Case I
			5.2.3.2 Power Control Optimization for Case II
			5.2.3.3 Feasibility of Problem (P2.1)
			5.2.3.4 Optimal Solution to Problem (P2.1)
		5.2.4 Simulation Results
			5.2.4.1 Simulation Setup and Benchmark Schemes
	5.3 Beamforming Design for MIMO AirComp FL
		5.3.1 Digital AirComp FL Model
			5.3.1.1 Digital Pre-processing at the Devices
			5.3.1.2 Post-processing at the PS
		5.3.2 Problem Formulation
		5.3.3 Optimization of Beamforming for FL Training Loss Minimization
			5.3.3.1 Analysis of the Convergence of the Designed FL
			5.3.3.2 Prediction of the Local FL Models
			5.3.3.3 Optimization of the Beamforming Matrices
		5.3.4 Simulation Results and Analysis
	5.4 Conclusions
6 Federated Learning for Autonomous Vehicles Control
	6.1 Autonomous Vehicle System Model
		6.1.1 Adaptive Longitudinal Controller Model
		6.1.2 FL Model
		6.1.3 Communication Model
	6.2 Dynamic Federated Proximal Algorithm for CAV Controller Design
		6.2.1 Dynamic Federated Proximal Algorithm
		6.2.2 Convergence of the DFP Algorithm
	6.3 Contract-Theory Based Incentive Mechanism Design
		6.3.1 Utility Function of the Parameter Server
		6.3.2 Utility Function of the CAVs
		6.3.3 Contract Design
	6.4 Simulation Results
	6.5 Conclusions
7 Federated Learning for Mobile Edge Computing
	7.1 MEC Network Model
		7.1.1 Transmission Model
		7.1.2 Computing Model
			7.1.2.1 Edge Computing Model
			7.1.2.2 Local Computing Model
		7.1.3 Time Consumption Model
		7.1.4 Energy Consumption Model
	7.2 Problem Formulation
	7.3 Federated Learning for Proactive User Association
		7.3.1 Components of the SVM-Based FL
		7.3.2 Training of SVM-Based FL
	7.4 Optimization of Service Sequence and Task Allocation
	7.5 Simulation Results and Analysis
	7.6 Conclusion
References