5.5 Deep Neural Networks
Deep networks stack many layers. Early layers spot simple stuff (edges), later layers spot complex stuff (faces).
Chapter 5: Deep Learning and Neural Networks
What makes a network "deep" is having many hidden layers - modern models can have dozens or even hundreds. Each layer builds on the last.
In an image model, early layers might detect edges and colours, middle layers detect shapes and textures, and later layers detect eyes, wheels or whole faces. It builds understanding from simple to complex - automatically.
Different jobs use different designs: CNNs (convolutional networks) are classic for images, RNNs handled sequences like text in the past, and Transformers now power most language and many image models.
Key takeaways
Deep = many hidden layers
Layers go from simple features to complex ones
CNNs for images, Transformers for language (and more)