HogWild++: A new mechanism for decentralized asynchronous stochastic gradient descent

Keyphrases

• Model Vectors 100%
• Asynchronous Stochastic Gradient Descent 100%
• Global Model 66%
• Stochastic Gradient Descent 66%
• Access Network 66%
• Non-uniform Memory Access 66%
• Multi-socket 66%
• Popular 33%
• Parallelizing 33%
• Main Idea 33%
• Local Model 33%
• Shared Memory 33%
• Remote Memory Access 33%
• Linear Speedup 33%
• False Sharing 33%
• Machine Learning Problems 33%
• Stochastic Gradient Algorithm 33%
• Multi-core Machines 33%
• Cache Invalidation 33%
• Token-based 33%
• Memory Access Conflict 33%
• Large-scale Machine Learning 33%
• Parallel Stochastic Gradient Descent 33%

Computer Science

• Gradient Descent 100%
• Non-Uniform Memory Access 33%
• Machine Learning 16%
• Multicore 16%
• Shared Memories 16%
• Learning Problem 16%
• Linear Speedup 16%
• Remote Memory Access 16%