RECYCLING MODEL UPDATES IN FEDERATED LEARNING: ARE GRADIENT SUBSPACES LOW-RANK?

Computer Science

• Centralized Model 25%
• Communication Overhead 25%
• Convergence Behavior 25%
• Deep Learning Model 25%
• Distributed Model 25%
• Distributed Systems 25%
• Experimental Result 25%
• Explained Variance 25%
• Federated Learning 100%
• Overwhelming Majority 25%
• Performance Model 25%
• Principal Components 25%

Keyphrases

• Distributed Model Training 25%
• Gradient Recycling 25%
• Gradient Space 50%
• Rank Characteristic 25%
• Subspace Low Rank 100%