PaSE: Parallelization Strategies for Efficient DNN Training

Venmugil Elango

Training a deep neural network (DNN) requires substantial computational and memory requirements. It is common to use multiple devices to train a DNN to reduce the overall training…