The human mind is able to identify objects in our vision, recognize words spoken in our hearing, understand multiple languages, and play games. Artificial intelligence (“AI”) is the discovery and study of methods to make computers mimic the functions of human minds.
To make computers perform useful functions, programmers typically write and test many thousands of lines of computer code, and frequently update that code over time. The code instructs a computer to accept inputs (like gestures on a touch screen or video from a camera) and generate outputs (like files, graphics on a display, or sounds from a speaker). Machine learning (ML) is the part of AI that makes computers learn new functions without writing code for each one.
In regression, the computer learns to generate a continuous-valued output from its input. For example, given inputs like the square footage of a home, the number of bedrooms, and its geographic location, the computer might learn to output the price of the home.
In classification, the computer learns to generate a discrete-valued output from its input. For example, given an e-mail as input, the computer might learn to classify it as “junk” or “not junk.”
In supervised learning, the computer is given training examples, each consisting of an input and the expected output. From these examples, the computer learns to generate the correct output, even for inputs it has never been given before.
In unsupervised learning, the computer is given example inputs, but no outputs. The computer learns to encode the input—represent it in some useful way. For example the output might be a compressed representation from which the input can be recovered. It might expose hidden similarities between inputs, allowing them to be clustered near each other. The output can also be used for detection of anomalies, hidden irregularities in inputs.