Feed-forward mapping networks

Abstract This chapter explores the ability of basic neural networks that feed input from and input layer through possible layers of hidden notes to the output layer. It is then shown that such neural networks can implement mapping functions. Mapping neural representations are important in many brain processes and have dominated models in cognitive science in the form of multilayer perceptrons. For this it is important to explore the effects of choosing appropriate values for synaptic weights through learning algorithms. While feedforward networks are not enough to explain cognitive functions alone, they are an important ingredient of brain-style information processing and have contributed greatly to our understanding of adaptive systems. This chapter includes some review about concepts of machine learning and more recent developments such as deep learning. Such techniques are becoming increasingly important for industrial applications as well as analysing neuroscience data. However, we will largely focus on their relation to brain-style information processing.