Explain what deep learning is, and what machine learning is in its supervised and unsupervised forms

Supervised learning trains on labelled data, where each observation has inputs and a known target, and learns to predict the target. A numerical target is a regression problem and a categorical one a classification problem. Unsupervised learning has no target and looks for structure in the inputs, reducing dimensions or grouping similar observations into clusters. Deep learning uses neural networks with many hidden layers, each transforming the previous layer's output through weights and non-linear activation functions, which suits images, speech and text. Reinforcement learning has an agent learn by trial and error to maximise a reward. The key distinction is whether a labelled target exists, not the type of data.