Kate Stevens critically evaluates the possibilities in a musical context where performance differences between trained and untrained listeners are modelled in sin gle and multilayer perceptrons and recurrent nets. She illustrates the use of measures such as response time and errors, and demonstrates how distributions of response time can be related graphically to precise locations of feature manipulations in musical stimuli. Additionally, she presents examples of the analysis of pitch and time based information in hidden unit space using the technique of canonical discriminant analysis.
Graeme Halford reviews how well neural networks account for the human reasoning involved in balance scale problems. He argues that human reasoning entails processing relational concepts, and, while these concepts can be represented by tensor products, this method involves weaknesses such as the need for hard-wiring and an inability to generalize. On the other hand, while three-layered backpropagation nets can, in principle, represent predicate-argument bindings for relations, they do not, as yet, model understanding of the balance scale in any comprehensive sense.
Simon Dennis examines the correspondence between psychological and model variables in human memory research including list length, repetition, presentation rate, item spacing, vocabulary, word frequency and recurrence frequency. Furthermore, he discusses methods of comparing measures of sensitivity (d'), reaction time and familiarity ratings recorded from subjects with estimates of the values of related model variables such as number of processing cycles and training epoch.