In this chapter, you learned how to use TensorFlow, an open source library for numerical computations with a special focus on deep learning. While TensorFlow is more inconvenient to use compared to NumPy, due to its additional complexity to support GPUs, it allows us to define and train large, multilayer neural networks very efficiently.

Also, you learned about the TensorFlow API to build complex machine learning and neural network models and run them efficiently. First, we explored programming in the low-level TensorFlow API. Implementing models at this level may be tedious when we have to program at the level of matrix-vector multiplications and define every detail of each operation. However, the advantage is that this allows us as developers to combine such basic operations and build more complex models. Furthermore, we discussed how TensorFlow allows us to utilize the GPUs for training and testing big neural networks to speed up the computations. Without the use of GPUs, training some networks would typically need months of computation!

We then explored two high-level APIs that make building neural network models a lot easier compared to the low-level API. Specifically, we used TensorFlow Layers and Keras to build the multilayer neural network and learned how to build models using those APIs.

Finally, you learned about different activation functions and understood their behaviors and applications. Specifically, in this chapter, we saw tanh, softmax, and ReLU. In Chapter 12, Implementing a Multilayer Artificial Neural Network from Scratch, we started with implementing a simple Multilayer Perceptron (MLP) to classify a handwritten image in the MNIST dataset. While the low-level implementation from scratch was helpful to illustrate the core concepts of a multilayer neural network, such as the forward pass and backpropagation, training neural networks using NumPy is very inefficient and impractical for large networks.

In the next chapter, we'll continue our journey and dive deeper into TensorFlow, and we'll find ourselves working with graph and session objects. Along the way, we'll learn many new concepts, such as placeholders, variables, and saving and restoring models in TensorFlow.