Section 2: Advanced Neural Network Architectures

This section familiarizes the reader with different types of convolutional and pooling layers that may be used in neural networks to process sensory input, from images on your laptop, to databases and real-time IoT applications. Readers will learn about using pretrained models, such as LeNet, and partial convolutional networks for image and video reconstruction on Keras, gain insights into how to deploy models using REST APIs, and then embed them in Raspberry computing devices for custom use cases, such as photography, surveillance, and inventory management.

Readers will be exposed to the underlying architectures of reinforcement learning networks in detail and learn how to implement core and extended layers in Keras for desired outcomes.

Then, they will dive deeper into the theory behind different types of recurrent networks and what it means to be a Turing-complete algorithm, examine specific consequences of backpropagation through time, including the issues of vanishing gradients, and obtain a comprehensive understanding of how temporal information is captured in such models. 

Readers will then explore a specific type of recurrent neural network (RNN) in detail, known as Long Short-Term Memory (LSTM), and become familiar with yet another neural network architecture that was inspired by our own biology.

This section comprises the following chapters:

• Chapter 4, Convolutional Neural Networks
• Chapter 5,  Recurrent Neural Networks 
• Chapter 6, Long-Short Term Memory Networks
• Chapter 7,  Reinforcement Learning with Deep Q-Networks