Matcher neural networks

In contrast to the prevailing ideas or bias regarding neural networks, the author takes a biologically conservative view that stresses efficiency and modularity. In this view individual neurons are efficient modules capable of biologically meaningful function. A group of neurons can form a larger module, which is also biologically meaningful and repeatable. Since modules are repeatable, the organization of the brain need not be as random as previously thought. With this new bias and the aim of discovering the modules, the author studies the physiology of neurons. He presents a powerful new class of neural networks, called matcher neural networks, based on the above ideas, which he believes are in many ways physiologically and psychologically better models for the brain than current connectionist models.<>