Training and Validation Plots

Training Hyper Parameters

We detail the hyper parameters of DNN models investigated in this paper as follows.

The Hyper Parameters for DNN Models on MNIST and CIFAR-10

The Hyper Parameters for DNN Models on IMDb

Note that the "LR" means learning rate, and "Tra.Para." indicates trainable parameters.

Training Performance Across Frameworks

The following figures show the training and validation plots of four models on GPU with different frameworks. Note that we train and validate LeNet-1 and LeNet-5 on MNIST, ResNet-20 and VGG-16 on CIFAR-10, and TextCNN, TextRNN(LSTM) and TextRNN(GRU) on IMDb.

LeNet-1

LeNet-5

ResNet-20

VGG-16

TextCNN

LSTM

GRU

Google Sites

Report abuse