Superlearning and neural network magic

Over the past five years neural networks have generated substantial interest in the application of complex adaptive systems to various fields of science and technology. It appears easy to raise enthusiasm, to fill numerous conferences each year, to start new journals, and to publish dozens of books and monographs. Researchers in fields such as psychology, physiology, biology, theoretical physics, mathematical statistics, artificial intelligence, computer science and electrical engineering as well as some people in pattern recognition have contributed to the area. It has been applied in many more areas. Only a small minority of all these scientists is able to demonstrate that they fully understand what is going on. For many it seems to be magic and this is probably part of the attraction. There are also people, however, with serious doubts as to the worth of what is gained by the neural network wave. I believe that there is a generally silent pattern recognition community who have such doubts. To support this statement, I will try to clarify the difference between some neural networks and traditional pattern recognition techniques. Finally, I will argue that there are certainly some things to do and to become less silent as either there is a not well understood phenomenon that should be studied, or neural network learning fits well in the traditional statistical pattern recognition par-