Layers

Introduction

Keras, a high-level neural networks API, is widely used for deep learning projects. It simplifies the process of building and training neural networks, making it accessible to both beginners and experts. In this chapter, we'll delve into the core of Keras: its layers. Layers are the fundamental building blocks of neural networks in Keras.

Note

Keras can be utilized as an independent library (from Keras 3.0 onwards) or as a component within TensorFlow or other deep learning frameworks. In this course, we will use it as a component within TensorFlow.

What Are Layers?

In the context of neural networks, a layer is a structured arrangement of neurons (nodes), typically organized in rows. Each neuron in a layer is connected to neurons in the previous and next layers. These connections are characterized by weights, which are adjusted during training.

Keras Layers

Keras, on the other hand, adopts a more modular and flexible approach. In Keras:

• Layers are more granular and specialized. Each layer typically performs a specific type of transformation or computation.
• Separation of concerns. Instead of a single layer handling both weighted sums and activations, Keras often separates these into different layer types. For instance, a Dense layer handles weighted sums, and an Activation layer handles the activation functions.

Input Layer

The Input layer is used to specify the shape of the input data.

Creating an Input Layer

Note

The Input layer doesn't process or modify data; it's a way to specify the input data shape.

Dense Layer

The Dense layer, also known as a fully connected layer, is a basic building block in neural networks. Each neuron in a dense layer receives input from all neurons of the previous layer, making it "fully connected."

Creating a Dense Layer

Here's how you create a single Dense layer in Keras:

Note

It's worth noting that the activation parameter can be utilized to directly incorporate an activation function following a Dense layer in your code. For example:

However, in many cases, it is advisable to treat the activation function as an independent layer. This approach offers greater flexibility in the design of your neural network.

Activation Layer

The Activation layer applies an activation function to its input. Common activation functions include 'relu', 'sigmoid', and 'tanh'.

Creating an Activation Layer

Note

The Activation layer independently applies an activation function to each individual input.