You have learned about PyTorch's core components, including Tensors, automatic differentiation with Autograd, defining models using torch.nn, and implementing data loading and training procedures. This chapter builds upon that foundation by introducing how to construct specific, widely used neural network architectures.

We will focus on two fundamental types of models:

1. Convolutional Neural Networks (CNNs): You'll understand the core concepts of convolutions and pooling, learn why CNNs are effective for grid-like data (especially images), and implement a basic CNN using layers such as nn.Conv2d and nn.MaxPool2d. We will also address how to manage the input and output shapes of these layers.
2. Recurrent Neural Networks (RNNs): You'll be introduced to the idea of processing sequential data using recurrent connections and hidden states. We will build a simple RNN using the nn.RNN layer and discuss the specific data format required for sequential inputs in PyTorch. A brief mention of more advanced variants like LSTMs and GRUs will also be included.

By the end of this chapter, you will be able to construct simple versions of these common architectures in PyTorch, preparing you to tackle more complex models later.