Machine Learning

A computer program normally does only what a human has explicitly told it to do, step by step. Machine Learning (ML) breaks that limitation. It is a subsystem of Artificial Intelligence in which computers acquire the ability to learn from data using statistical techniques, without being explicitly programmed by a human being.

At the heart of Machine Learning are algorithms that use data to learn on their own and make predictions. Instead of a programmer encoding every rule, the algorithm discovers patterns in the data it is given and builds its own basis for predicting what comes next.

These learning algorithms are called models, and they pass through a definite life cycle before they are trusted with real work:

• Training — the model is first trained using a set of data called the training data. This is where it learns.
• Testing — the model is then tested using a separate set called the testing data, which checks how well what it learned holds up on data it did not learn from.
• Successive trainings — the cycle is repeated, refining the model each time.
• Prediction — once, after these successive trainings, the model gives results at an acceptable level of accuracy, it is put to use making predictions about new and unknown data.

The order matters and is a favourite examination point: a model is first trained (on training data), then tested (on testing data), and only after repeated trainings bring its accuracy to an acceptable level is it used on genuinely unseen data. …