Neural computers. (Proceedings of the NATO advanced research workshop on neural computers held in Neuss, Germany, F. R., September 28-October 2, 1987) (Q4039737)

The book brings 50 contributions, 4 group reports (on structure, software simulation, hardware implementation and application of neural nets, respectively), extensive separate bibliography, author and subject indices; due to the broad scope of covered topics, this makes the publication a sort of a reference book. Several lectures present competent overviews and insights on general philosophy, paradigmata and impact of neural networks, written by leading experts (T. Kohonen, J. A. Feldman, C. v. d. Malsburg, W. R. Moore, E. Bienenstock, W. v. Seelen and H. A. Mallot, T. J. Sejnowski, A. J. Pellionisz, R. Hecht-Nielsen). The bulk of special contributions is centered around adaptive learning theory, vision and motor modelling, and implementation. The first area is represented by various generalizations, classifications and cation area and target systems, software quality features etc.) If one develops a program using a transformation system, a design decision corresponds to the choice of a transformation rule. This choice is a control of program development on a meta-level. The connection between non-functional properties and design decisions is described by the use of logic. An inference mechanism helps choosing a transformation rule. A classical inference mechanism is not sufficient because the knowledge on the connection between non-functional properties and design decisions is uncertain and incomplete. The use of many-valued and non-monotonic logics leads to more adequate results.