The machine learning ( machine-learning in English) is one of the fields of study of the Artificial intelligence.

Machine learning refers to the development, the analysis and the implementation of methods which make it possible a machine (in the broad sense) to evolve/move thanks to a process of training, and thus to fill of the tasks which it is difficult or impossible to fill by more traditional algorithmic means.

Here two examples of applications of machine learning:

• One can conceive a machine system of learning allowing a robot, having the capacity to move its members but not knowing anything the coordination of the movements allowing walk, to learn how to go. The robot will start by carrying out movement random, then, by privileging the movements enabling him to advance, will set up little by little an increasingly effective walk.
• the Character recognition is a complex task because two similar characters are never exactly equal. One can conceive a machine system of learning which learns has to recognize characters by observing examples, i.e. known characters.

It is trying to take as a starting point the living beings to design machines able to learn. Thus, even if machine learning is before a whole under-field of the Informatique, it is also closely related to the cognitive Sciences, with the Neuroscience S, the Biologie and the Psychologie.

Types of training

The algorithms of training can be categorized according to the type of training which they employ:

• the Training supervised: an expert (or oracle) is employed to label examples correctly. Learning must then find or approximate the function which makes it possible to assign the good label to these examples. The linear discriminating Analysis or SVM is typical examples.
• the Training not-supervised (or automatic classification): No expert is available. The algorithm must discover by itself the structure of the data. The clustering and the model of mixtures of Gaussian are not supervised algorithms of training.
• the Training by reinforcement: the algorithm learns a behavior being given an observation. The action of the algorithm on the environment produces a value of return which guides the algorithm of training. The algorithm of Q-learning east is a traditional example.

The algorithms which one generally meets in this field are:

• the machines with vector of support

• the Booster pump
• the networks of neurons for a training supervised or not-supervised
• the Method of K closer neighbors for a training supervised
• the decision trees
• the Statistical methods as for example the Model of Gaussian mixture
• the logistic Regression
• the linear discriminating Analysis
• the fuzzy Logical
• the genetic algorithms and the genetic Programming

These methods are often combined to obtain various alternatives of training. The use of such or such algorithm strongly depends on the task to solve (classification, estimate of values, etc). Machine learning is used in a very broad spectrum of applications: Search engine, assistance with the diagnosis, Bio-data processing, detection of frauds, analyzes financial markets, Voice recognition, the manuscript writing, analyzes and indexing of images and video, Robotique…

Random links:

Roquebrune (Gers) | The Middle Ages with the cinema | Sabine Karsenti | Bartsia alpina | Camaro Yenko