Deep Neural Evolution by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9789811536854
Publisher: Springer Singapore

Fig. 8.4The average convergence graph of the GA population in one representative run on SVHN dataset

Table 8.3 shows the structures of the best networks evolved in different GA runs for each dataset. Each row of Table 8.3 shows the CNN architecture in terms of the number of blocks, number of convolutional layers in each block, and the feature map size of each block. The convolutional blocks are separated by comma and in each block we show the number of layers and the feature map size (in parenthesis) for each layer in that block. For example in the first row of Table 8.3, 4 × (128) represents that the first convolutional block of the CNN consists of 4 convolutional layer each having a feature map size of 128. The complete CNN model is constructed by adding ReLU, batch normalization, dropout, average pooling layer and fully connected layers as described in Sect. 8.4.1. This table also shows the number of trainable parameters for different evolved models. The other hyperparameters of each evolved network for CIFAR10, CIFAR100, and SVHN dataset are shown in Tables 8.4, 8.5, and 8.6, respectively. Figure 8.5 visualizes the architecture of the best CNN models evolved by GA over five repeated runs in three datasets.

Fig. 8.5Architectures of top CNN models evolved by GA (left to right for: CIFAR10, CIFAR100, and SVHN datasets)

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.