Enhancing Distributed Neural Network Training Through Node-Based Communications
S Moreno-Álvarez, ME Paoletti… - IEEE transactions on …, 2023 - ieeexplore.ieee.org
The amount of data needed to effectively train modern deep neural architectures has grown
significantly, leading to increased computational requirements. These intensive …
significantly, leading to increased computational requirements. These intensive …
Training deep neural networks: a static load balancing approach
Deep neural networks are currently trained under data-parallel setups on high-performance
computing (HPC) platforms, so that a replica of the full model is charged to each …
computing (HPC) platforms, so that a replica of the full model is charged to each …
Equilibrado de carga dirigido por modelos de Kernels de datos paralelos en plataformas heterogéneas de alto rendimiento
S Moreno Álvarez - 2019 - dehesa.unex.es
Data-Parallel applications are composed of several processes that apply the same
computation (kernel) to different amounts of data. While its execution, these applications …
computation (kernel) to different amounts of data. While its execution, these applications …