How Neural Networks Work
Understand how neural networks learn, from neurons and backpropagation to convolutional and recurrent networks, then build a self-driving car from scratch.
Start learning Open course bookAbout this course
Neural networks sit behind most modern AI, but their core ideas are approachable: weighted sums, activation functions, a measure of error and a method for reducing it. Once those are clear, the famous architectures, convolutional networks for images and recurrent networks for sequences, follow naturally.
This course follows three freeCodeCamp courses. The first explains how deep neural networks work, with the mathematics made visual. The second is a code-free crash course in the key terms and ideas of deep learning, from loss functions and optimisers to the five steps of building any model. The third puts it all into practice by building a self-driving car simulation in plain JavaScript, with a neural network trained by a genetic algorithm. Each lesson comes with an original book chapter of worked examples.
What you will learn
- Explain how neurons, weights, biases and activation functions combine into layers
- Work through a backpropagation step by hand using the chain rule
- Describe how convolutional and recurrent networks process images and sequences
- Choose sensible loss functions, optimisers and regularisation for a problem
- Build and train a small neural network from scratch with a genetic algorithm
Who this course is for
Beginners who want a real understanding of neural networks. The first two lessons need no coding; the third uses JavaScript. School algebra is enough, and the chapters explain every calculation they use.