Communication-Efficient Federated DNN Training: Convert, Compress, Correct

Computer Science

• Deep Neural Network 100%
• Neural Network Training 100%
• Quantisation Error 50%
• Parameter Server 50%
• Normal Distribution 25%
• Neural Network Model 25%
• Communication Overhead 25%
• Mathematical Convergence 25%
• Gradient Descent 25%
• Compression Scheme 25%
• Floating Point 25%
• Learning Rate 25%
• Lossless Compression 25%

Mathematics

• Deep Neural Network 100%
• Quantisation Error 50%
• Stochastics 25%
• Gaussian Distribution 25%
• Nonuniform 25%
• Convergence Analysis 25%
• Network Model 25%
• Lossless Compression 25%

Keyphrases

• Memory Decay 66%
• Gradient Quantization 33%