Neural Network Models

Abstract

Models of large groups of neurons can be very diverse. In some models individual parameters of real life neurons can still be distinguished. In the models of Little (1974) and Shaw and Vasudevan (1974) one can still speak about a membrane potential and the individual action potentials. Simplifications and generalizations occur on both sides. One extreme is to consider only the average firing rate of a neuron as conveying information (e.g., Sejnowski 1976 a), another extreme is to consider neurons as logical switching elements (e.g., McCulloch and Pitts 1943). It is interesting to note, however, that in the limit of large numbers of neurons both extremes are capable of representing the same neural net. We will discuss a limited sample of neural network models at this point and will start with the logical neuron models, then advance with more realistic Little-type neurons and finally treat some examples of continuous modeling of neural nets. In later chapters we will discuss more stereotyped network models which deal either exclusively with a particular brain structure or with particular properties such as plasticity, learning and associative memory.