Patent №
US 3,934,231
Granted
—
Owner
—
Lab
—
AI components
3
ml · kr · hardware
Assignment
None on record
Dataset
AIPD
2023_r1 edition
Application
05446968
A digital logic circuit, hereinafter referred to as an element, of which an interconnected plurality together with their connections to a control unit comprise a learning machine which synthesizes a boolean function of n variables as the result of a training procedure. Each element may operate as a two-input, one-input combinational circuit of one of four logical types, where the type is determined by the current internal state of the element. All four types are such that a (ZERO,ZERO) input pair gives rise to a ZERO output, and a (ONE,ONE) input pair gives rise to a ONE output, while the particular operation realized is determined by two function value units which compute suitable outputs under the (ONE,ZERO) and (ZERO,ONE) input pairs to the element. Three different species of element are described differing in their computation of heuristic responsibility. The "global search" species is useful when the function to be synthesized is not constant; there is a convergence theorem related to this species. The "latest error" species is also useful for applications where the function to be synthesized may be a constant; but this type does not obey a known convergence theorem. The "hill climbing" species is useful principally to improve an existing approximate synthesis. Means are provided for setting and reading out the functions realized by the elements in a tree-like network in order to facilitate storage and transmission of synthesized functions.