Turing's Connectionism: An Investigation of Neural Network Architectures

Rating:
Summary: Turing's neural networks and genetical search
Review: A programmable digital computer, based on Alan Turing's design, ran its first program in 1950 at the National Physical Laboratory, London. Even today, every computer in the world remains computationally equivalent to a Turing Machine. It is little known however, that Turing also investigated neural network architectures as early as 1948, and before the term genetic algorithm was coined, proposed configuring his networks with a "genetical search". In this book Teuscher presents the most extensive exploration of Turing's neural networks available. The book contains over 100 diagrams, detailed examinations of the logical behaviour of Turing's networks, experiments into their emergent properties and extensions of Turing's ideas based on recent findings. An understanding of Turing's networks allows insight into a number of modern research areas such as Kauffman's work on the principles of self-organisation, the boundaries of computability, and even the real neural networks of living things (Turing claimed that his neural networks were probably the simplest possible model of the human cortex). Because the discussion in the book starts with Turing's early networks and progresses through to current research, it can also be read as an accessible overview of the history of the field. In addition, the book makes it clear that there are many interesting research questions still to be answered in this area. As such, this book will be of interest to historians of computer science and modern researchers alike.