This book aims to help readers have a systematic understanding of deep learning technology through practical systems and develop their own strategies on network design. To achieve this goal, the book adopts a diagnostic and prescriptive approach. The book starts with breaking down a canonical deep learning network into blocks and layers to understand the complexity and behavior of the network, bottlenecks and issues are identified as a result.

A series of advanced network engineering methods are presented targeting specific issues in deep learning design. Those methods include recurrent convolutional neural network, residual convolutional neural networks, 1x1 transformation, autoencoder, U-nets, graph convolution network, region-based convolutional neural networks, YOLO object detection network, backpropagation and generative adversarial networks.