Learning and generalization in logic trees

Summary form only given, as follows. Results were obtained on the learned synthesis of Boolean functions using tree networks whose elements, after training, perform logical operations AND and OR on two or more inputs. The tree acts on a Boolean input vector, and its complements. Generalization results from the insensitivity of the binary tree functions to changes of their input vectors. The concept of parsimonious evaluation, derived from the property of AND and OR to have determined outputs when only one input is known, was shown to lead to significant speedups both in software and in hardware implementations