Video by freeCodeCamp.org · taught by Jason Dsouza · streamed from YouTube. All rights remain with the creator.
Lesson 2: Deep Learning Key Concepts
About this lesson
This crash course by Jason Dsouza introduces deep learning to absolute beginners using diagrams and plain explanations, without any code, so that the key ideas and vocabulary are clear before you start programming. It begins by explaining what deep learning is and how it fits within artificial intelligence and machine learning, then introduces neural networks and the way they learn by passing data forwards and errors backwards. A long section then works through the core terms that appear everywhere in the field. It explains activation functions and why networks need them, loss functions and how they measure a model's mistakes, and optimisers, the methods that decide how the weights change during training. It clarifies the difference between parameters, which a model learns, and hyperparameters, which you choose, and explains what epochs, batches and iterations mean. The course next surveys the three main kinds of learning, supervised, unsupervised and reinforcement learning, with examples of each, and explains regularisation, the group of techniques that stop a network from simply memorising its training data. An introduction to network architectures follows, covering fully connected feedforward networks, recurrent networks for sequences and convolutional networks for images. The course ends with five steps that apply to every deep learning project, from gathering data to optimising the trained model. Afterwards, read Chapter 2 of the course book, which works softmax and loss examples by hand.