Cognitive Science An Introduction to the Science of the Mind

This chapter considers how connectionist neural networks offer a contrast to the symbolic view of representation discussed in previous chapters. We start by reviewing the structure of neural networks inspired by neurobiology, comparing a single unit in a neural network to a biological neuron. The second section looks at the simplest form of neural network -- a single-layer neural network using the perceptron convergence rule for learning. The third section introduces multilayer neural networks and the development of the backpropagation algorithm. Next, we look at how the multilayer neural network can be trained, and its biological plausibility. The last section summarizes three critical features of information processing in neural networks, as opposed to physical symbol systems: distributed representations, the lack of a clear distinction between storing and processing information, and the ability to learn.