KEYWORDS: evolutionary neural networks, adaptive systems, non-stationary environment, neural networks, evolutionary computation

In several real-world dynamical systems applications, such as controlling robots, or planning a strategy in a multiplayer game, or dynamically allocating a time-dependant resource, it’s not possible to extract an a priori model of the system taking into account all the variables influencing the evolution during time because of the presence of some unobservable dynamics in the system. In such cases a possible approach is to extract a first rough offline model, and then to on line update it, in order to recover the gap of knowledge about unknown disturbances. Trying to pursue this on line model adaptation has been the research subject of many groups involved in the development of smart adaptive systems. The problem has been classified as an on line modelling task and the one of the most commonly ascribed approach to reach that goal consists in matching the neural networks modelisation capabilities with the adaptation properties of evolutionary algorithms. These investigations led to the birth of a new framework for adaptive systems which is generally referred to as Evolutionary Artificial Neural Networks (EANNs) where the modelling potentialities of artificial neural networks have been matched with the adaptation properties of the evolutionary algorithms. The model we get from a neural network undergoes the evolution superimposed by the artificial environment of the evolutionary algorithm, which is in turn related to the evolution of the system. In such a way it’s possible to couple the evolution of the population and the learning process of each individual of the population, achieving better adaptation of the whole environment to a generic dynamic fitness landscape.

In this paper we present the state-of-the-art of the topic and review the current theoretical and methodological approaches to develop evolutionary neural network. In particular we focus our attention on those works where a continuous on line adaptation of the model to a non stationary environment is required. It’s commonly thought that this is one of the most troubling features in real-world applications and, even if the first theoretical approaches to the problem in terms of adaptive artificial systems can be fixed in the first 90’s, the real implementations started later and at present they are not consolidated yet. Preliminary applications overviewed in this paper show good adaptivity to unpredictable variations in time of the dynamics of the system and the results obtained can be considered both a promise and a challenge for future research in smart adaptive systems.