FedNL: Making Newton-Type Methods Applicable to Federated Learning

Keyphrases

• Hessian 100%
• Newton's Method 100%
• Federated Learning 100%
• Training Data 50%
• Learning Techniques 50%
• Contractive 50%
• Uplink 25%
• Numerical Experiments 25%
• Line Search 25%
• Generalized Linear Model 25%
• Communication Complexity 25%
• Second-order Methods 25%
• Communication Efficiency 25%
• Cubic Regularization 25%
• Art Communication 25%
• Model Compression 25%
• Compression Operators 25%
• Local Convergence Rate 25%
• Partial Participation 25%
• Error Feedback 25%
• Bidirectional Compression 25%

Computer Science

• Federated Learning 100%
• Learning Technique 100%
• Condition Number 50%
• Communication Complexity 50%
• Model Compression 50%
• Training Data Point 50%
• Regularization 50%
• Convergence Rate 50%
• Training Data 50%