MIPD: An Adaptive Gradient Sparsification Framework for Distributed DNNs Training

Computer Science

• Deep Neural Network 100%
• Sparsification 100%
• Interpretability 33%
• Probability 33%
• Deep Learning Method 16%
• Large Data Set 16%
• Exponential Smoothing 16%
• Network Usage 16%
• Server Architecture 16%
• Parameter Server 16%
• Experimental Result 16%
• Gradient Magnitude 16%
• Error Gradient 16%
• Learning System 16%
• Graphics Processing Unit 16%

Keyphrases

• Gradient Quantization 50%
• Distributed Deep Learning System 25%