Trustworthy Federated Learning: First International Workshop, FL 2022 Held in Conjunction with IJCAI 2022 Vienna, Austria, July 23, 2022 Revised Selected Papers

Description

This book constitutes the refereed proceedings of the First International Workshop, FL 2022, Held in Conjunction with IJCAI 2022, held in Vienna, Austria, during July 23-25, 2022.  The 11 full papers presented in this book were carefully reviewed and selected from 12 submissions. They are organized in three topical sections: answer set programming; adaptive expert models for personalization in federated learning and privacy-preserving federated cross-domain social recommendation. Preface Organization Contents Adaptive Expert Models for Federated Learning 1 Introduction 2 Background 2.1 Problem Formulation 2.2 Regimes of Non-IID Data 2.3 Federated Learning 2.4 Iterative Federated Clustering 2.5 Federated Learning Using a Mixture of Experts 3 Adaptive Expert Models for Personalization 3.1 Framework Overview and Motivation 4 Experiments 4.1 Datasets 4.2 Non-IID Sampling 4.3 Model Architecture 4.4 Hyperparameter Tuning 4.5 Results 5 Related Work 6 Discussion 7 Conclusion References Federated Learning with GAN-Based Data Synthesis for Non-IID Clients 1 Instruction 2 Related Works 3 Preliminary 4 Synthetic Data Aided Federated Learning (SDA-FL) 5 Experiments 5.1 Experiment Setup 5.2 Evaluation Results 6 Conclusions and Discussions References Practical and Secure Federated Recommendation with Personalized Mask 1 Introduction 2 Preliminaries 2.1 Matrix Factorization 2.2 Federated Matrix Factorization 3 Federated Masked Matrix Factorization 3.1 Personalized Mask 3.2 Adaptive Secure Aggregation 4 Experiments 4.1 Settings 4.2 Efficiency Promotion and Privacy Discussion 4.3 Discussion on Model Effectiveness 5 Conclusion References A General Theory for Client Sampling in Federated Learning 1 Introduction 2 Background 2.1 Aggregating Clients Local Updates 2.2 Unbiased Data Agnostic Client Samplings 2.3 Advanced Client Sampling Techniques 3 Convergence Guarantees 3.1 Asymptotic FL Convergence with Respect to Client Sampling 3.2 Application to Current Client Sampling Schemes 4 Experiments on Real Data 5 Conclusion References Decentralized Adaptive Clustering of Deep Nets is Beneficial for Client Collaboration 1 Introduction 2 Related Work 3 Method 3.1 Non-IID Data 3.2 DAC: Decentralized Adaptive Clustering 3.3 Variable DAC 4 Experimental Setup 5 Results on Covariate Shift 6 Results on Label Shift 7 Conclusions References Sketch to Skip and Select: Communication Efficient Federated Learning Using Locality Sensitive Hashing 1 Introduction 2 Related Work 3 Methods 3.1 Sketch-Based Communication Skipping: Sketch-to-Skip 3.2 Sketch-Based Client Selection: Sketch-to-Select 3.3 Sketch to Skip and Select FL Algorithm 4 Experiments 4.1 Experimental Setup 4.2 Results 5 Conclusions References Fast Server Learning Rate Tuning for Coded Federated Dropout 1 Introduction 2 Background 3 Methodology 3.1 Fast Server Learning Rate Adaptation 3.2 Coded Federated Dropout 4 Evaluation 5 Conclusion and Future Works References FedAUXfdp: Differentially Private One-Shot Federated Distillation 1 Introduction 2 Related Work 3 FedAUX 3.1 Method 3.2 Privacy 4 FedAUXfdp 4.1 Regularized Empirical Risk Minimization 4.2 Privacy 4.3 Cumulative Privacy Loss 5 Experiments 6 Conclusion References Secure Forward Aggregation for Vertical Federated Neural Networks 1 Introduction 2 Introduction 2.1 Background: SplitNN in VFL 2.2 Trade-Off of SplitNN 3 Secure Forward Aggregation 3.1 Overview 3.2 Aggregation Method 3.3 Training with Weight Mask 3.4 Removable Mask on Transformed Data 3.5 Security Analysis 3.6 Mitigate Trade-Off Using SFA 4 Experiment 4.1 Experiment Setting 4.2 Dataset 4.3 Performance of SFA 4.4 Trade-Off Between Security and Model Performance 5 Related Work 6 Conclusion References Two-Phased Federated Learning with Clustering and Personalization for Natural Gas Load Forecasting 1 Introduction 2 Related Work 2.1 Time-Series Forecasting 2.2 Federated Learning 2.3 Personalized Federated Learning 3 Method 3.1 Knowledge-Based Federated Clustering 3.2 Two-Phase Federated Learning 3.3 Attention-Based Model Aggregation Strategy 4 Case Studies 4.1 Experimental Settings 4.2 Forecasting Performance 4.3 Forecasting Performance Distribution 4.4 Performance on Different Number of Clusters 5 Conclusion and Future Work References Privacy-Preserving Federated Cross-Domain Social Recommendation 1 Introduction 2 Preliminaries 2.1 Social Recommendation System 2.2 Vertical Federated Learning 2.3 Differential Privacy 3 Problem Formulation 3.1 System Model 3.2 Threat Model 4 The Scheme of Federated Cross-Domain Social Recommendation 4.1 The Algorithm Framework 4.2 Learning of Social Networks with Differential Privacy 4.3 Highly Efficient SG"0365G Construction and Calculation for U'c 4.4 Privacy-Preserving User Feature Vector Update with Matrix Confusion Method 5 Security Analysis 6 Experiments 6.1 Experimental Setting 6.2 Evaluation on Training Effect 6.3 Comparison with Existing Federated Schemes 7 Conclusions and Future Works References Author Index