Machine learning is a subfield of AI (Artificial Intelligence).
The aim of machine learning is to automatically recognize laws, patterns or procedures from existing data material. In image processing, for example, objects within an image are to be recognized.
In the classical approach, an algorithm or program code is explicitly implemented by a programmer who specifies how the input data is to be processed. Once the logic has been defined, it is more or less fix and can only be changed by manually adjusting the algorithm.
In machine learning, an attempt is made to map dependencies in the data using a dynamic model. The concrete parameterization of the model however is not determined by a programmer. Instead, in a training phase, training data is applied to a learning algorithm in order to automatically find an optimal setting for the model parameters. The goal and at the same time the challenge is, to find a model setting which maps both the patterns in the training data but even more important, unknown data which has not been shown to the training algorithm (generalization).
Furthermore, machine learning is characterized by the fact that this training phase is not done once and never changed any more. It is repeated at certain intervals in order to adapt the model to changing data.
See also supervised learning.