Topics: The following topics are covered in this module

1. Neural Network Neurons, their functions and uses
2. Weights and biases
3. The training process
4. The Perceptron

Objectives

By the end of this module, students will be able to:

1. Describe the basis of a neural network (neuron).
2. Identify and describe an artificial neuron (perceptron).
3. Discuss bias and weights.
4. Describe and identify activation functions.
5. Describe and simulate image processing in a small neural network.
6. Implement and train a perceptron using TensorFlow.

Watch

Video: Deep Learning Foundations: Opening the Machine Video

Topics

• Neural Network Neurons
• Neuron Function
• Where Do Weights and Bias Come From?
• The Training Process
• Model Deployment and Neurons

Exercise

Coding a perceptron

The perceptron is relatively simple—only one neuron to worry about! If this simple network seems complicated to code, don’t worry! We are doing most of the coding from scratch in this exercise. This will help you see how neurons work and follow individual components through the training process. After this notebook though, we’ll leave the details to the AI frameworks and use APIs that make all of this much easier!

Work through notebook 02.1_code_a_perceptron.ipynb.

Network Capacity

Network capacity refers to the complexity or flexibility of a neural network to capture patterns in data. A network with higher capacity can model more intricate relationships but is also more prone to overfitting, meaning it might perform well on training data but poorly on unseen data. Factors affecting capacity include:

• The number of layers
• The number of neurons in each layer
• The type of activation functions used

Image Processing

Due to the magic of universal approximation, the tools that let us program a perceptron can be expanded to recognize images! In the following Jupyter Notebook, we’ll go more in-depth with how to create a model from layers and train it!

Work through notebook 02.2_mnist_classifier.ipynb.