- 84 Downloads
This chapter deals with the information representation in neural networks and the description of the information content of several types of neurons and networks using the concept of sigma-algebra. The main idea is to describe the evolution of the information content through the layers of a network. The network’s input is considered to be a random variable, being characterized by a certain information. Consequently, all network layer activations will be random variables carrying forward some subset of the input information, which are described by some sigma-fields. From this point of view, neural networks can be interpreted as information processors.