Pattern recognition in a non-stationary environment
Dynamical Learning in non-stationary environments

When data evolve over time, it is important that the classifier is able to adapt to these changes. These changes may be due either to a normal variation in the system's parameters and/or structure, or to degradations affecting its intrinsic characteristics and behavior. These systems, called evolutionary, dynamic or non-stationary, cover a wide range of applications: electromechanical, chemical, thermal, medical, energy, etc. They manipulate non-stationary data, which can present three major difficulties. The first difficulty lies in the enormous quantity of data, constantly generated by various applications over time. These data streams are potentially infinite, and it is almost impossible to store them. Only a brief summary of this data can be extracted and stored for future use. The second difficulty is due to the high arrival speed of these data streams, which prevents them from being...