Promise of Neural Networks

The neural network is one of the most exciting new tools for information processing to have emerged over the last few years. An increasing number of computer tasks require the use of human-like capabilities, a fact which has caused severe problems for many industries and services in the past. Now, however, neural networks are being developed which can handle a broad range of tasks in vision, speed recognition, time series analysis, robotics and control. They have proved particularly useful in dealing with problems which cannot be solved by explicit rules, which involve noisy data, and which allow a system to be trained by examples. In response to increasing neural network activity during the 1980s, the DEANNA Project (Database for European Artificial Neural Network Activity) was established in 1990 (ESPRIT Project P5668). The main objective of this 15-month EC-funded project was to construct a database of current European neural computing research which could be used to assess the success of various applications, and to pinpoint the major difficulties in technology transfer. The project also aimed to create a European neural network advisory structure, and to develop neural computing simulators and demonstrators. This text contains a comprehensive overview of neural networks based on the findings of the DEANNA Project, together with a view of the broader framework of the subject in a world context. The first part of the report describes neural nets and their available hardware and software simulators. The second part discusses deeper principles, and reviews a range of applications. The final section describes the results of the DEANNA questionnaire, and provides a summary of developments in current European neural network activity.