Information Geometry on Complexity and Infomax Principles

Using the information-geometric framework, it is possible to naturally define the complexity of a composite system as its deviation from the set of non-complex systems. Here, a system is considered to be non-complex if it is just the superposition of its constituent elements. The aim of my presentation is to give some analytical results on complex systems and to illustrate them by computer simulations. Applying my approach to the field of neural networks, it leads to a generalized version of Linsker's infomax principle.