Basically, developers distinguish with progressive intermediate steps supervised learning) and unsupervised learning. The algorithms used are very different. Supervised learning brings examples, like a database, into the system. Developers specify the value of the information, for example, whether it belongs to category A or B. The machine learning system draws conclusions, recognizes patterns and can better handle unknown data. The goal is to further reduce the error rate.
A known example of supervised learning is spam filter: the system uses functions to decide whether mail is sent to the inbox or placed in the spam (junk mail) folder. If the system makes an error, you can correct it manually and the filter will adjust its future calculations. The software therefore obtains better results. Such a filtering program is based on the set of Bayesian filters (from probability theory, Bayes' theorem) which is why we speak of Bayesian spam filtering.
THE unsupervised learningthat's to say unsupervised learningeliminates the teacher, who in supervised learning always indicates what belongs and provides feedback on the system's autonomous decisions. Instead, the program here tries to recognize patterns on its own. It can use clustering (partitioning of data), for example: an element is selected from the quantity of data, examined for its characteristics and then compared to those already examined. If it has already examined equivalent elements, the current object will be added. If not, then it is stored separately.
Systems based on unsupervised learning are implemented in neural networks. Application examples can be found in the network security : a machine learning system detects abnormal behavior. For example, since a cyberattack cannot be attributed to a known group, the program can then detect the threat and notify a problem, alarming the user.
In addition to these two main areas, there is also thesemi-supervised learning (semi-supervised learning) And reinforcement learning (reinforcement learning), And active learning (active learning): These three methods are more closely related to supervised learning and differ in the type and extent of user participation.
Furthermore, there is also a distinction between shallow learning and the deep learning. While the first method is relatively simple, with fairly superficial results, deep learning is more difficult to understand. This is very complex information, because it involves natural informationfor example those that occur during speaking, writing or facial recognition. Natural data is easy for humans to process, but not for a machine, because it is difficult to grasp mathematically.
Deep learning and an artificial neural network are closely linked. The way a neural network is trained can be described as deep learning. We call it deep learning because the neural network is organized into several hierarchical levels. The first level begins with a layer of input neurons. They record the data, begin their analysis and send their results to the next neural node. At the end, the increasingly refined information reaches the initial level and the network delivers a value. The sometimes very numerous levels located between the entrance and the exit are called hidden layers (hidden layers).
