Deep Learning and its Applications using Python

8Advanced Deep Learning

An advanced different models of deep learning algorithm are used frequents. An up-gradation or modifications are possible because of the flexible nature of neural network which lead to design an end to end model. So this advancement allows a researcher to build simple to complex structure based on the need aligned with imagination. Some of the advances deep learning algorithms [1] AlexNet, VGG, NiN, GoogLeNet, ResNet, DenseNet, GRU, LSTM, D-RNN, and Bi-RNN. These advanced deep learning algorithms are discussed in detail along with its code for implementation using pythons Keras library in TensorFlow platform.

8.1 Deep Convolutional Neural Networks (AlexNet)

It holds the structure of convolutional neural network [2] which had a spotlight on image net large-scale visual recognition challenge in 2012. It has eight layers – five layers with max pooling layer and three with fully connected layer. It this Relu activation function have been used in all fully connected layers other than output layer by which the performance of the model is improved six times than before. Which overcome the over fitting by using dropout layer. The architecture of AlexNet is shown in Figure 8.1.

Consider an example of training the model with ImageNet dataset which has millions of images with thousands of classes. To process this input deep model have been introduced where it perform padding to save the feature maps from reducing its size. The representation of layers are shown ...