Artificial Neural Networks: The Origin of the Species at BCL

Pei Ling Lai
Department of Electronic Engineering
Southern Taiwan University
Tainan, Taiwan

Alfred Inselberg
School of Mathematical Sciences
Tel Aviv University,
Tel Aviv, Israel

McCulloch and Pitts (1943) at MIT developed models of neural networks which computed simple logic functions (i.e. "or", "and"). Rosenblatt (1958) at Cornell University attracted considerable interest in the Perceptron which could "learn" to associate a given input to a desired output. Minsky and Papert's book (1969) exposed the limitations of the Perceptron and its extensions, and nearly eliminated funding for research in neural networks as a result. Various researchers persisted and during the late 1970s and early 1980s interest in neural network re-emerged and since then the field has flourished. In 1959, starting from the McCulloch Pitts networks, "Property Filters" – neural networks -- were developed at BCL to do contour detection, pattern recognition and much more. BCL's remarkable achievements are not widely known though they enjoy frequent rediscovery. But why did BCL's successful approach experience an unnatural evolutionary selection – omission – from the literature of modern artificial neural networks?