SPATL: Salient Parameter Aggregation and Transfer Learning for Heterogeneous Federated Learning

Keyphrases

• Transfer Learning 100%
• Federated Learning 100%
• Parameter Transfer 100%
• Parameter Aggregation 100%
• Heterogeneous Federated Learning 100%
• Communication Cost 66%
• Communication Overhead 66%
• Training Process 66%
• Model Inference 33%
• Edge Devices 33%
• Control Mechanism 33%
• Limited Resources 33%
• State-of-the-art Techniques 33%
• Parameter Selection 33%
• Agents of Selection 33%
• Heterogeneous Clients 33%
• Resource Heterogeneity 33%
• Data Heterogeneity 33%
• Gradient Control 33%
• U-Net 33%
• Stable Training 33%
• User Data Privacy 33%
• Shared Encoder 33%
• VGG11 33%
• Local Predictor 33%
• Local Inference 33%
• Learning Methods 33%
• AI Models 33%
• Model Convergence 33%

Computer Science

• Transfer Learning 100%
• Federated machine learning 100%
• Communication Cost 50%
• Communication Overhead 50%
• Training Process 50%
• Speed-up 25%
• User Data 25%
• Data Privacy 25%
• Inference Model 25%
• Data Heterogeneity 25%
• Residual Neural Network 25%
• Control Mechanism 25%
• Artificial Intelligence 25%