Book Image

Mastering TensorFlow 1.x

Book Image

Mastering TensorFlow 1.x

Overview of this book

TensorFlow is the most popular numerical computation library built from the ground up for distributed, cloud, and mobile environments. TensorFlow represents the data as tensors and the computation as graphs. This book is a comprehensive guide that lets you explore the advanced features of TensorFlow 1.x. Gain insight into TensorFlow Core, Keras, TF Estimators, TFLearn, TF Slim, Pretty Tensor, and Sonnet. Leverage the power of TensorFlow and Keras to build deep learning models, using concepts such as transfer learning, generative adversarial networks, and deep reinforcement learning. Throughout the book, you will obtain hands-on experience with varied datasets, such as MNIST, CIFAR-10, PTB, text8, and COCO-Images. You will learn the advanced features of TensorFlow1.x, such as distributed TensorFlow with TF Clusters, deploy production models with TensorFlow Serving, and build and deploy TensorFlow models for mobile and embedded devices on Android and iOS platforms. You will see how to call TensorFlow and Keras API within the R statistical software, and learn the required techniques for debugging when the TensorFlow API-based code does not work as expected. The book helps you obtain in-depth knowledge of TensorFlow, making you the go-to person for solving artificial intelligence problems. By the end of this guide, you will have mastered the offerings of TensorFlow and Keras, and gained the skills you need to build smarter, faster, and efficient machine learning and deep learning systems.
Table of Contents (21 chapters)
19
Tensor Processing Units

Understanding pooling

Generally, in the convolution operation several different kernels are applied that result in generation of several feature maps. Thus, the convolution operation results in generating a large sized dataset.

As an example, applying a kernel of shape 3 x 3 x 1 to an MNIST dataset that has images of shape 28 x 28 x 1 pixels, produces a feature map of shape 26 x 26 x 1. If we apply 32 such filters in a convolutional layer, then the output will be of shape 32 x 26 x 26 x 1, that is, 32 feature maps of shape 26 x 26 x 1.

This is a huge dataset as compared to the original dataset of shape 28 x 28 x 1. Thus, to simplify the learning for the next layer, we apply the concept of pooling.

Pooling refers to calculating the aggregate statistic over the regions of the convolved feature space. Two most popular aggregate statistics are the maximum and the average. The output...