Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training

INIS

• symmetry 100%
• neural networks 100%
• matrices 42%
• calculation methods 42%
• algorithms 42%
• neurons 42%
• operation 28%
• graphs 14%
• datasets 14%
• reduction 14%
• speech 14%
• simulation 14%
• range 14%
• losses 14%
• resources 14%
• accuracy 14%
• accelerators 14%
• devices 14%
• learning 14%

Computer Science

• Recurrent Neural Network 100%
• Neural Network Training 100%
• Sparsity 42%
• training algorithm 28%
• Hardware 28%
• Model 14%
• Logic Simulation 14%
• Gradient Computation 14%
• Memory Access 14%
• Long Short-Term Memory Network 14%
• Propagation Phase 14%
• Computation Graph 14%
• Computation Matrix 14%
• Matrix Operation 14%
• Hardware Accelerator 14%
• Arithmetic Operation 14%
• Computing Resource 14%
• Temporal Sequence 14%
• Algorithms 14%
• Accuracy 14%
• Matrix Multiplication 14%

Keyphrases

• Limited Computing Resources 14%
• Fluent Speech 14%
• Online Incremental Learning 14%
• Sequence Task 14%
• Symmetric Computation 14%
• Dense Matrix multiplication 14%
• Logic Simulation 14%