Forms of machine learning

There are several forms of machine learning. In one case, the computer is presented with inputs, and together with each one it is given a response to that input. This is called supervised learning. The desired response could be to assign the input to one of a fixed number of classes (like the letters of the alphabet). This is called classification. In other situations, the desired response is a real number (such as the likely profit from selling a product in a certain location). The latter case is called regression.

Many people use spreadsheet programs to perform linear regression. Linear regression requires choosing a small set of basic models for the answer and combining them in a simple way. Supervised learning doesn't require choosing such models, so it allows more flexibility in determining the responses to the inputs.

The rule, or program, that determines for each possible input an appropriate response is called a function. Another kind of learning is reinforcement learning, where after performing sequences of actions, some feedback is given as to how satisfactory the result of the sequence of actions was. This is called a reinforcement. Humans learn this way from birth, before there is any symbolic way of communicating to the child what the correct response to a certain situation is. Reinforcements take the form of food, comfort, a soothing tone, freedom from pain and later on more symbolic ones: verbal appreciation, prizes and -- this should come as no surprise -- money.