Communication-Efficient Local Stochastic Gradient Descent for Scalable Deep Learning

Computer Science

• Gradient Descent 100%
• Simulation Mode 85%
• Algorithms 42%
• Communication 42%
• Parallelism 28%
• Model Accuracy 28%
• Data Parallelism 28%
• Deep Learning 28%
• Accuracy 28%
• Experimental Result 14%
• Scalability 14%
• Parameter Model 14%
• Convergence Rate 14%
• Communication Cost 14%
• Design 14%

INIS

• stochastic processes 100%
• workers 71%
• accuracy 57%
• algorithms 42%
• periodicity 42%
• data 28%
• learning 28%
• design 14%
• cost 14%
• losses 14%
• trains 14%
• metrics 14%
• convergence 14%