Picture for Andrew C. Kirby

Andrew C. Kirby

Layer-Parallel Training with GPU Concurrency of Deep Residual Neural Networks Via Nonlinear Multigrid

Add code
Jul 14, 2020
Andrew C. Kirby, Siddharth Samsi, Michael Jones, Albert Reuther, Jeremy Kepner, Vijay Gadepally
Figure 1 for Layer-Parallel Training with GPU Concurrency of Deep Residual Neural Networks Via Nonlinear Multigrid
Figure 2 for Layer-Parallel Training with GPU Concurrency of Deep Residual Neural Networks Via Nonlinear Multigrid
Figure 3 for Layer-Parallel Training with GPU Concurrency of Deep Residual Neural Networks Via Nonlinear Multigrid
Figure 4 for Layer-Parallel Training with GPU Concurrency of Deep Residual Neural Networks Via Nonlinear Multigrid

A Multigrid Full Approximation Storage algorithm for solving Deep Residual Networks is developed to enable neural network parallelized layer-wise training and concurrent computational kernel execution on GPUs. This work demonstrates a 10.2x speedup over traditional layer-wise model parallelism techniques using the same number of compute units.

* 7 pages, 6 figures, submitted to 2020 IEEE High Performance Extreme Computing Conference 
Viaarxiv icon
Github IconAccess Paper or Ask Questions
Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon